{"id":"dad8c08e-2170-4bc8-b0de-6574375f0e89","entityType":"agent","slug":"clawhub-jaccen-3dgs-method-compare","name":"3dgs Method Compare","canonicalUrl":"https://www.xpersona.co/agent/clawhub-jaccen-3dgs-method-compare","canonicalPath":"/agent/clawhub-jaccen-3dgs-method-compare","generatedAt":"2026-10-11T17:42:18.993Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T14:39:21.758Z","emptyReason":null},"description":"Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories. Skill: 3dgs Method Compare Owner: jaccen Summary: Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories. Tags: latest:1.4.7 Version history: v1.4.7 | 2026-05-19T07:37:17.438Z | auto - Expanded built-in knowledge base to 523+ methods and 24 categories (up from 254+ and 21). - Added a new comparison table for rendering formulations, including new primitive","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s172m233k07zcmx035knhdv0nh85v0ts:3dgs-method-compare","sourceUrl":"https://clawhub.ai/jaccen/3dgs-method-compare","homepage":"https://clawhub.ai/jaccen/skills/3dgs-method-compare","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/jaccen/3dgs-method-compare","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/jaccen/skills/3dgs-method-compare","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":60,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories. Skill: 3dgs Method Compare Owner: jaccen "},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T14:39:21.758Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T14:39:21.758Z","emptyReason":null},"stars":null,"forks":null,"downloads":1048,"packageName":null,"latestVersion":"1.4.7","tractionLabel":"1K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T14:39:21.694Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T14:39:21.758Z","lastCrawledAt":"2026-10-11T14:39:21.694Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T14:39:21.694Z","lastVerifiedAt":null,"highlights":[{"version":"1.4.7","createdAt":"2026-05-19T07:37:17.438Z","changelog":"- Expanded built-in knowledge base to 523+ methods and 24 categories (up from 254+ and 21). - Added a new comparison table for rendering formulations, including new primitives and compositing techniques. - Included newly published methods such as SNS (Skew-Normal Splatting) and MGS. - Updated tables and method descriptions to reflect newly tracked developments. - Improved classification and indexing across all method types for easier comparison.","fileCount":3,"zipByteSize":11120},{"version":"1.4.4","createdAt":"2026-05-16T10:59:39.116Z","changelog":"**Expanded database and method categories, new primitives, and more detailed comparison support.** - Increased built-in knowledge from 150+ to 254+ 3DGS methods - Includes 21 method categories (up from previous) - Added new primitive type: spatially-varying (SVGS) - Updated compression and geometry/surface methods with additional recent techniques - Comparison dimensions and tables reflect latest method variants and features","fileCount":2,"zipByteSize":8803},{"version":"0.1.2","createdAt":"2026-05-06T01:08:22.285Z","changelog":"- Expanded built-in knowledge base: now covers 150+ 3D Gaussian Splatting (3DGS) methods (up from 105+). - Added more recent methods in all categories, including foundation, robustness, geometry/surface, compression, language/semantic, and feed-forward approaches. - Included new comparison entries for methods such as Softmax-GS, LeGS, GETA-3DGS, Luminance-GS++, 2D-SuGaR, IRIS, DiffSoup, ArtifactWorld, GLMap, and NG-GS. - Updated and extended technical tables for deeper, up-to-date comparisons. - Version incremented to 1.3.0.","fileCount":2,"zipByteSize":6769},{"version":"0.1.1","createdAt":"2026-04-30T10:45:50.682Z","changelog":"- Increased built-in knowledge base from 104+ to 105+ 3DGS methods. - Updated description and documentation to reflect the expanded method coverage. - No code changes; documentation and database adjustment only.","fileCount":2,"zipByteSize":5058},{"version":"0.1.0","createdAt":"2026-04-30T07:56:28.508Z","changelog":"**Initial public release of 3dgs-method-compare.** - Enables detailed, multi-dimensional comparison of 3D Gaussian Splatting (3DGS) variants, covering 104+ methods. - Generates publication-quality tables analyzing primitives, opacity, color, rendering, performance, and more. - Includes method database with foundation, compression, geometry, language, feed-forward, and SLAM variants. - Supports recommendations and trade-offs based on specific research or application scenarios. - Accepts comparison triggers in both English and Chinese.","fileCount":2,"zipByteSize":5005}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s172m233k07zcmx035knhdv0nh85v0ts:3dgs-method-compare","setupComplexity":"low","setupSteps":["Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T17:42:18.991Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-jaccen-3dgs-method-compare/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-11T14:39:21.758Z","emptyReason":null},"readme":"Skill: 3dgs Method Compare\n\nOwner: jaccen\n\nSummary: Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories.\n\nTags: latest:1.4.7\n\nVersion history:\n\nv1.4.7 | 2026-05-19T07:37:17.438Z | auto\n\n- Expanded built-in knowledge base to 523+ methods and 24 categories (up from 254+ and 21).\n- Added a new comparison table for rendering formulations, including new primitives and compositing techniques.\n- Included newly published methods such as SNS (Skew-Normal Splatting) and MGS.\n- Updated tables and method descriptions to reflect newly tracked developments.\n- Improved classification and indexing across all method types for easier comparison.\n\nv1.4.4 | 2026-05-16T10:59:39.116Z | auto\n\n**Expanded database and method categories, new primitives, and more detailed comparison support.**\n\n- Increased built-in knowledge from 150+ to 254+ 3DGS methods\n- Includes 21 method categories (up from previous)\n- Added new primitive type: spatially-varying (SVGS)\n- Updated compression and geometry/surface methods with additional recent techniques\n- Comparison dimensions and tables reflect latest method variants and features\n\nv0.1.2 | 2026-05-06T01:08:22.285Z | auto\n\n- Expanded built-in knowledge base: now covers 150+ 3D Gaussian Splatting (3DGS) methods (up from 105+).\n- Added more recent methods in all categories, including foundation, robustness, geometry/surface, compression, language/semantic, and feed-forward approaches.\n- Included new comparison entries for methods such as Softmax-GS, LeGS, GETA-3DGS, Luminance-GS++, 2D-SuGaR, IRIS, DiffSoup, ArtifactWorld, GLMap, and NG-GS.\n- Updated and extended technical tables for deeper, up-to-date comparisons.\n- Version incremented to 1.3.0.\n\nv0.1.1 | 2026-04-30T10:45:50.682Z | auto\n\n- Increased built-in knowledge base from 104+ to 105+ 3DGS methods.\n- Updated description and documentation to reflect the expanded method coverage.\n- No code changes; documentation and database adjustment only.\n\nv0.1.0 | 2026-04-30T07:56:28.508Z | auto\n\n**Initial public release of 3dgs-method-compare.**\n\n- Enables detailed, multi-dimensional comparison of 3D Gaussian Splatting (3DGS) variants, covering 104+ methods.\n- Generates publication-quality tables analyzing primitives, opacity, color, rendering, performance, and more.\n- Includes method database with foundation, compression, geometry, language, feed-forward, and SLAM variants.\n- Supports recommendations and trade-offs based on specific research or application scenarios.\n- Accepts comparison triggers in both English and Chinese.\n\nArchive index:\n\nArchive v1.4.7: 3 files, 11120 bytes\n\nFiles: skill-card.md (1999b), SKILL.md (25231b), _meta.json (138b)\n\nFile v1.4.7:SKILL.md\n\n---\r\nname: 3dgs-method-compare\r\ndescription: \"Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories.\"\r\nversion: 1.4.7\r\nauthor: jaccen\r\ntags: [\"3dgs\", \"gaussian-splatting\", \"method-comparison\", \"research\"]\r\n---\r\n\r\n# 3DGS Method Comparison Engine\r\n\r\nYou are an expert in 3D Gaussian Splatting methods with deep knowledge of 523+ variants. Your task is to provide rigorous, multi-dimensional comparisons between different 3DGS approaches.\r\n\r\n## Capabilities\r\n\r\n- Compare any combination of 3DGS variants across 10+ technical dimensions\r\n- Generate publication-quality comparison tables\r\n- Analyze design trade-offs and identify positioning\r\n- Provide recommendation based on specific use cases\r\n\r\n## Comparison Dimensions\r\n\r\nWhen comparing methods, analyze across the following dimensions:\r\n\r\n### 1. Primitive Representation\r\n- Shape: Full 3D Gaussian / 2D disk / 1D splat / hybrid / spatially-varying (SVGS)\r\n- Anisotropy: Isotropic / Anisotropic / Semi-anisotropic\r\n- Parameterization: (μ, Σ, opacity, SH) / (center, normal, scale, opacity) / custom / (μ, Σ, spatially-varying color+opacity, SH) (SVGS)\r\n\r\n### 2. Opacity / Alpha Mechanism\r\n- Range: [0, 1] / [-1, 1] / unbounded / sigmoid / tanh\r\n- Signed support: Yes (signed α) / No (standard GS)\r\n- Negative mechanism: Negative color (NegGS) / Negative opacity (signed) / None\r\n\r\n### 3. Color Representation\r\n- Spherical Harmonics order: 0/1/2/3\r\n- Color space: RGB / HDR / Feature vectors\r\n- Negative color support: Yes (NegGS) / No\r\n\r\n### 4. Rendering Formulation\r\n- Rasterization: Tile-based / Forward / Deferred\r\n- Blending: Front-to-back / Back-to-front\r\n- Anti-aliasing: EWA splatting / Mip-aware / None\r\n\r\n### 5. Frequency & Geometry Modeling\r\n- High-frequency boundary: Explicit / Implicit / None\r\n- Surface quality: Point-based / Surfels / Hybrid\r\n- Geometric constraints: Depth normal / ESDF / Mesh prior\r\n\r\n### 6. Density Control\r\n- Strategy: Clone + Split + Prune / Progressive / Anchor-based\r\n- Adaptivity: Gradient-based / Loss-based / Statistics-based\r\n- Compression: Pruning / Quantization / Distillation\r\n\r\n### 7. Training Strategy\r\n- Resolution schedule: Coarse-to-fine / Fixed\r\n- Iterations: 7k / 30k / custom\r\n- Regularization: Depth / Normal / Smoothness / Sparsity\r\n\r\n### 8. Performance Characteristics\r\n- Speed (FPS): Real-time (>30) / Interactive (10-30) / Offline (<10)\r\n- Memory: VRAM requirement\r\n- Storage: Model size (MB)\r\n- Scalability: Small object / Room-scale / City-scale\r\n\r\n### 9. Applicable Scenarios\r\n- Novel view synthesis\r\n- Surface reconstruction\r\n- 3D editing\r\n- Dynamic scenes\r\n- Large-scale scenes\r\n- Autonomous driving\r\n\r\n### 10. Code & Reproducibility\r\n- Official implementation available\r\n- Framework: PyTorch / JAX / CUDA / Custom\r\n- Dependencies\r\n\r\n## Rendering Formulation Comparison\r\n\r\n| Method | Primitive | Compositing | Key Feature |\r\n|--------|-----------|-------------|-------------|\r\n| 3DGS | 3D Anisotropic Gaussian | alpha-compositing (front-to-back) | Tile-based rasterization |\r\n| Softmax-GS | 3D Anisotropic Gaussian | Softmax competition | Replaces α-compositing with learnable softmax |\r\n| Mip-Splatting | 3D Anisotropic Gaussian + Mip | alpha-compositing | 3D smoothing + 2D Mip filter |\r\n| 3DGEER | 3D Anisotropic Gaussian | Exact ray-Gaussian integral | Replaces splatting with exact rendering |\r\n| SNS | Azzalini Skew-Normal Distribution | alpha-compositing | Learnable skewness for asymmetric boundaries |\r\n\r\n## Known Methods Database\r\n\r\n### Foundation Methods\r\n\r\n| Method | Venue | Primitive | Opacity | Key Feature |\r\n|--------|-------|-----------|---------|-------------|\r\n| 3DGS | SIGGRAPH'23 | 3D anisotropic | [0,1] sigmoid | Tile-based rasterization |\r\n| Mip-Splatting | CVPR'24 (Best Student Paper) | 3D anisotropic + Mip | [0,1] | 3D smoothing + 2D Mip filter, alias-free |\r\n| 2DGS | SIGGRAPH'24 | 2D disk | [0,1] | Better surface reconstruction |\r\n| Scaffold-GS | ICCV'23 | Anchor+3D | [0,1] | Anchor-based scalability |\r\n| Scaffold-GS+ | CVPR'24 | Anchor+3D | [0,1] | Progressive training |\r\n| Softmax-GS | CVPR'26 (Findings) | 3D anisotropic | Softmax competition | Replaces α-compositing with learnable softmax; blend-vs-bound |\r\n| LeGS | arXiv'26 | 3D anisotropic | RL-controlled | RL-based learnable density control replacing heuristics; O(N) reward |\r\n| SNS | arXiv'26 (2605.15010) | Skew-Normal | [0,1] | Skew-Normal primitive replacing symmetric Gaussian kernels; continuous interpolation between symmetric Gaussian ↔ Half-Gaussian via learnable skewness |\r\n\r\n### Signed / Decomposed Methods\r\n\r\n| Method | Opacity Range | Color Range | Mechanism |\r\n|--------|--------------|-------------|-----------|\r\n| NegGS | [0, +∞) (non-negative) | ℝ (negative allowed) | Negative color + Diff-Gaussian |\r\n| (Standard GS) | [0, 1] via sigmoid | [0, +∞) | Standard α-compositing |\r\n\r\n**Critical Distinction**: Methods using \"negative\" concepts differ fundamentally:\r\n- **Signed opacity (α ∈ [-1,1])**: Opacity α can be negative, rendering formula modified. The Gaussian primitive itself carries a sign. Better for sharp geometric boundaries.\r\n- **NegGS**: Opacity remains non-negative, but color values can be negative. Uses Diff-Gaussian (subtraction of two Gaussians) to model ring/crescent structures.\r\n\r\n### Compression Methods\r\n\r\n| Method | Compression Ratio | Quality Impact | Speed |\r\n|--------|-------------------|----------------|-------|\r\n| Compact-3DGS | 10-15x | Minimal PSNR drop | Faster |\r\n| LightGS | 15-20x | Slight drop | Much faster |\r\n| MobileGS | 50-100x | Moderate drop | Real-time mobile |\r\n| Embedded-3DGS | 10x | Minimal | Comparable |\r\n| HAC | ~100x | Slight drop | Faster after decode |\r\n| OT-UVGS | UV tensor | ↑ vs spherical UVGS | Same as UVGS |\r\n| NanoGS | Training-free | Minimal (KNN merge) | CPU-only, instant |\r\n| MesonGS++ | 34x | Minimal | Faster after decode (0-1 ILP hyperparameter search) |\r\n| GETA-3DGS | 5x | Minimal | First end-to-end automatic joint structured pruning + quantization; QADG; render-aware saliency |\r\n| CAGS | ~7x (streaming) | Minimal | VQ-based compression with Level-of-Detail streaming; progressive decode for bandwidth-adaptive deployment |\r\n| MGS | arXiv'26 (2603.19234) | Any LoD prefix | Matryoshka continuous LoD via stochastic budget training; renders any prefix k splats |\r\n\r\n### Robustness / Regularization Methods\r\n\r\n| Method | Venue | Prior Source | Key Feature |\r\n|--------|-------|-------------|-------------|\r\n| EnerGS | arXiv'26 | LiDAR (partial geometric) | Energy-based soft guidance instead of hard constraints; improves outdoor large-scale scenes |\r\n| Luminance-GS++ | TPAMI'26 | Illumination prior | Illumination-robust NVS; decouples shading from geometry |\r\n\r\n### Geometry / Surface Methods\r\n\r\n| Method | Venue | Surface Quality | Key Feature |\r\n|--------|-------|----------------|-------------|\r\n| 2DGS | SIGGRAPH'24 | High | Oriented 2D disks for geometry |\r\n| SuGaR | CVPR'24 | High | Surface-aligned regularization |\r\n| PGSR | TVCG'24 | Highest (SOTA) | Planar regularizer + unbiased depth rendering |\r\n| PAGaS | arXiv'26 | High (depth) | 1DoF Gaussians for depth refinement |\r\n| Vol3DGS | CVPR'25 | High | Volume-consistent rendering |\r\n| 2D-SuGaR | arXiv'26 | Highest (DTU SOTA) | 2DGS + monocular depth/normal priors; depth-guided init; clustering-based pruning |\r\n| IRIS | arXiv'26 (2603.15368) | Hybrid | GS-proxy neural field with analytical ray intersection; hybrid rendering |\r\n| DiffSoup | arXiv'26 (2603.27151) | Extreme simplification | Triangle soup as alternative primitive to Gaussians |\r\n| 3DSS | arXiv'26 (2605.05876) | High (inverse rendering) | First differentiable surface splatting; coverage-based compositing from EWA; joint shape+SVBRDF+lighting |\r\n| SVGS | arXiv'24 (2411.18966) | High (Blender SOTA) | Spatially varying color+opacity within each Gaussian; movable kernels (1.4x params); >30 FPS |\r\n| AmbiSuR | ICML'26 | High (photometric) | Photometric ambiguity disambiguation for accurate GS surface reconstruction |\r\n| DySurface | arXiv'26 | High (4D surface) | Bridges explicit Gaussians and implicit SDF for consistent 4D surface reconstruction |\r\n\r\n### Generation / Text-to-3D\r\n\r\n| Method | Venue | Input | Output | Key Feature |\r\n|--------|-------|-------|--------|-------------|\r\n| DreamGaussian | ICLR'24 (Oral) | Text prompt | 3D mesh + 3DGS | SDS + 3DGS prior, seconds |\r\n| GaussianEditor | Preprint | Text/geometry mask | Edited 3DGS | CLIP-guided selection + editing |\r\n| ArtifactWorld | arXiv'26 (2604.12251) | Artifact images | Restored video | Video generation for artifact restoration |\r\n| SceneGen-LLMRL | arXiv'26 (2605.05711) | Language | Interactive 3D scene | LLM-RL coupling for unified 3D scene generation + immersive interaction |\r\n\r\n### Language / Semantic\r\n\r\n| Method | Venue | Feature Source | 3D Storage | Key Feature |\r\n|--------|-------|---------------|------------|-------------|\r\n| LangSplat | CVPR'24 | CLIP (2D distillation) | Per-Gaussian CLIP features | Open-vocabulary 3D queries |\r\n| Feature 3DGS | CVPR'24 | DINO/SAM (2D distillation) | Per-Gaussian feature vectors | Downstream task features |\r\n| NRGS | arXiv'26 | Neural network | Learned regularization | Robust semantic 3DGS |\r\n| Semantic Foam | CVPR'26 (Highlight) | Volumetric Voronoi mesh | Per-cell semantic feature field | Semantic decomposition; outperforms Gaussian Grouping, SAGA |\r\n| GLMap | CVPR'26 | Multi-scale semantics | Per-Gaussian language features | Gaussian-Language Map; zero-shot navigation |\r\n| NG-GS | arXiv'26 (2604.14706) | NeRF-guided | Per-Gaussian segmentation | NeRF-guided GS segmentation |\r\n| PointGS | CVPR'26 | SAM masks (contrastive distillation) | Per-Gaussian semantic features | 3DGS as unified intermediate for unsupervised 3D point cloud segmentation; SAM→3D contrastive learning |\r\n\r\n### Feed-Forward Methods\r\n\r\n| Method | Venue | #Gaussians | Inference | Key Feature |\r\n|--------|-------|------------|-----------|-------------|\r\n| GlobalSplat | Preprint'26 | ~16K | <78ms | Global scene tokens, 4MB footprint |\r\n| MVSplat | ECCV'24 | Variable | Single-pass | Cost-volume-based prediction |\r\n| GS-LRM | ECCV'24 | Variable | Single-pass | 1B transformer, zero-shot generalization |\r\n| DepthSplat | CVPR'25 | Variable | Single-pass | Stereo-guided depth regularization |\r\n| InstantSplat | arXiv'24 | Variable | ~40s total | Pose-free sparse-view |\r\n| AnySplat | SIGGRAPH'25 | Variable | Single-pass | In-the-wild unconstrained views |\r\n| SparseSplat | CVPR'26 | 22% of SOTA | Single-pass | Pixel-unaligned, entropy-based probabilistic sampling, 3D-Local Attribute Predictor |\r\n| OT-UVGS | EG'26 | UV tensor | Same as UVGS | OT-based UV mapping, O(N log N) |\r\n| Free Geometry | arXiv'26 | Adaptive | Single-pass + LoRA | Self-evolving feed-forward, +3.73% camera accuracy |\r\n| FTSplat | arXiv'26 (2603.05932) | Variable | Single-pass | Feed-forward triangle splatting |\r\n| SplatWeaver | arXiv'26 (2605.07287) | Variable | Single-pass | Cardinality Gaussian Expert Routing (Null/1/2/3 experts per pixel) + DWT frequency prior; 30% Gaussian budget with +1.02 dB PSNR over AnySplat |\r\n\r\n### SLAM Methods\r\n\r\n| Method | Venue | Input | Scale | Key Feature |\r\n|--------|-------|-------|-------|-------------|\r\n| Gaussian Splatting SLAM | CVPR'24 (Highlight) | Monocular video | Room-scale | First real-time monocular 3DGS SLAM, differentiable rendering for joint pose+map |\r\n| CGS-SLAM | IROS'25 | Monocular video | Room-scale | Voxel-based compact representation for efficiency |\r\n| WildGS-SLAM | CVPR'25 | Monocular video | Room-scale | Dynamic environments, uncertainty-aware mapping via pretrained 3D priors |\r\n| S3PO-GS | ICCV'25 | Monocular video | Outdoor | Scale-consistent pose optimization, eliminates outdoor scale drift |\r\n| Flow4DGS-SLAM | arXiv'26 | Monocular video | Room-scale | Optical flow-guided 4DGS for temporal consistency |\r\n| E2EGS | CVPR'26 (2603.14684) | Event camera | Room-scale | Event-camera pose-free 3D reconstruction |\r\n| MAGS-SLAM | arXiv'26 | RGB (multi-agent) | Multi-room | First RGB-only multi-agent 3DGS SLAM; compact submap communication + geometry/appearance-aware loop verification |\r\n\r\n### Large-Scale Methods\r\n\r\n| Method | Venue | Scale | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| Scaffold-GS | ICCV'23 | Building | Anchor-based efficiency |\r\n| Scaffold-GS+ | CVPR'24 | City | Progressive training |\r\n| CityGaussian | ECCV'24 | City | Hierarchical LOD |\r\n| Street Gaussians | ECCV'24 | Street | Static/dynamic decomposition, driving scenes |\r\n| Octree-GS | Preprint | City | Octree acceleration + LOD |\r\n\r\n### Cross-Domain Applications\r\n\r\n| Method | Venue | Domain | Key Feature |\r\n|--------|-------|--------|-------------|\r\n| GS-DOT | arXiv'26 | Medical (DOT) | Diffusion transport for photon imaging |\r\n| BiSplat-WRF | IEEE ICC'26 Workshop | Wireless (WRF) | Planar GS + bilinear spatial transformer for EM coupling |\r\n| FieryGS | ICLR'26 | Physics simulation | Physics-integrated fire synthesis |\r\n| SplAttN | ICML'26 (Spotlight) | Point cloud completion | Gaussian soft splatting for point cloud completion |\r\n| Fake3DGS | ICPR'26 | Forensics | First benchmark for 3D manipulation detection in neural rendering |\r\n| SandSim | arXiv'26 | Digital art | Curve-guided Gaussian for sand painting reconstruction |\r\n| RGS | arXiv'26 | Medical (CBCT) | Residual wavelet-GS for sparse-view CBCT |\r\n| RESPIRE | arXiv'26 | Medical (bronchoscopy) | CT-informed mesh-anchored GS for dynamic bronchoscopy |\r\n| Color-Encoded Illumination | CVPR'26 (Highlight) | High-speed imaging | Color-coded temporal info for volumetric reconstruction |\r\n| HDR-NSFF | ICLR'26 (2603.08313) | Dynamic HDR scenes | HDR dynamic scene neural scene flow fields |\r\n| 3DGS AD Safety Eval | SafeComp'26 | Autonomous driving | Industrial fidelity evaluation for AD perception |\r\n| HeroGS | CVPR'26 | Sparse-view NVS | Hierarchical guidance for sparse-view robust 3DGS |\r\n| Sparse-View 3DGS Wild | arXiv'26 | Sparse-view NVS | Diffusion-guided sparse-view enhancement |\r\n| Pi-GS | arXiv'26 (2602.03327) | Sparse-view NVS | Sparse-view with π³ reference-free initialization |\r\n| GS-Surrogate | arXiv'26 (2604.06358) | Physics simulation | Deformable GS for simulation visualization |\r\n| 3DGEER | ICLR'26 | Rendering (exact) | Exact ray-Gaussian rendering replacing splatting; fisheye/generic camera support; top 1% |\r\n| Forecast-GS | arXiv'26 | Robotics | Predictive GS for forecasting task-completed states in robotic manipulation |\r\n| GaussianGrasper | T-RO'24 | Robotics / Grasping | Open-vocabulary grasping via SAM+CLIP feature distillation into 3DGS |\r\n| GraspSplats | CoRL'24 | Robotics / Grasping | Zero-shot manipulation with 3D feature splatting; scene editing support |\r\n| ManiGaussian | ECCV'24 | Robotics / Manipulation | Dynamic GS world model for multi-task manipulation via future scene prediction |\r\n| GSMem | arXiv'26 | Embodied Reasoning | 3DGS as persistent spatial memory for zero-shot embodied exploration & QA |\r\n| RoboSplat | RSS'25 | Robotics / Data Gen | Diverse data generation via Gaussian primitive manipulation; 87.8% success |\r\n| VR-Robo | RAL'25 | Robotics / Navigation | Real-to-Sim-to-Real for visual robot navigation without depth sensors |\r\n| GSDrive | arXiv'26 | Driving RL | 3DGS environment for reinforcing driving policies |\r\n| GeoQuery | SIGGRAPH'26 | Sparse-view NVS | Geometry-guided cross-view attention with geometry-aligned proxy queries from predicted depth |\r\n| PairDropGS | arXiv'26 | Sparse-view NVS | Paired dropout-induced consistency regularization with progressive scheduling |\r\n| VidSplat | SIGGRAPH'26 | Sparse-view NVS | Training-free generative framework leveraging video diffusion priors with iterative confidence refinement |\r\n| OCH3R | arXiv'26 (2605.13018) | Single RGB | Object-Centric Holistic 3D from single RGB; per-pixel CLIP + 6D pose + per-object Gaussians |\r\n\r\n### Dynamic / 4DGS Methods\r\n\r\n| Method | Venue | Primitive | Rendering | Key Feature |\r\n|--------|-------|-----------|-----------|-------------|\r\n| FreeTimeGS++ | arXiv'26 (2605.03337) | 4D Gaussians + durations | Gated marginalization | Neural velocity fields + emergent temporal partitioning; comprehensive 4DGS analysis |\r\n| ParticleGS | arXiv'26 | 3D anisotropic + physics | Standard α-compositing | Physics-based motion extrapolation for fluid/dynamic scenes; Lagrangian particle dynamics |\r\n| TransmissiveGS | arXiv'26 | Dual-GS (surface + reflection) | Deferred shading | Transmissive + reflective dual decomposition; separate G-buffer compositing for glass/refractive objects |\r\n| PD-4DGS | arXiv'26 | 3-layer progressive (static + global deform + local refine) | Progressive streaming | DASH/HLS-compatible 4DGS streaming; ~1.7s first-frame latency vs 73-930s monolithic |\r\n| 3DGS³ | arXiv'26 | 3D anisotropic (super-sampled) | Standard + temporal interpolation | Gradient-Aware Super Sampling + Lightweight Temporal Frame Interpolation for large-scale 3DGS |\r\n| BlitzGS | arXiv'26 | 3D anisotropic (distributed) | Parity-based multi-GPU | Distributed city-scale GS training; parity-based sharding across multi-GPU; eliminates single-GPU memory bottleneck |\r\n| Z-Order GS | arXiv'26 | 3D anisotropic (Z-ordered) | Z-order curve indexing | Z-order curve spatial indexing for cache-coherent Gaussian traversal; improved rendering throughput |\r\n| PanoPlane | arXiv'26 | Planar (panoramic) | Plane-based compositing | Panoramic plane-based GS for omnidirectional NVS; efficient panoramic scene representation |\r\n| SparseOIT | arXiv'26 | 3D anisotropic | Order-independent transparency | Sparse order-independent transparency for correct See-through rendering of overlapping semi-transparent Gaussians |\r\n| SCOUP | arXiv'26 | Sparse code primitives | Language-conditioned | Sparse code language GS; language-conditioned sparse coding for controllable 3DGS generation |\r\n| AV1-3DGS | arXiv'26 | 3D anisotropic | AV1 motion-vector SfM | AV1 codec motion vectors for dense SfM; 63% training time reduction; leverages video compression priors |\r\n| RoSplat | arXiv'26 | 3D anisotropic (feed-forward) | Pixel-wise GS | Feed-forward pixel-wise GS for sparse-view NVS; requires alpha normalization for varying view counts |\r\n| HarmoGS | arXiv'26 | 3D anisotropic | Harmonized optimization | Gradient harmonization for in-the-wild 3DGS; resolves cross-view gradient conflicts from transient distractors and illumination inconsistencies |\r\n| GuardMarkGS | arXiv'26 | 3D anisotropic | Watermark + deterrence | First unified watermarking + edit deterrence framework for 3DGS assets; security for 3D content |\r\n| FaceParts | arXiv'26 | 3D anisotropic (part-based) | Part-compositional | Part-based decomposable Gaussian avatar; modular facial region modeling for expressive avatars |\r\n| RetroNVS | arXiv'26 | 3D anisotropic | Retro-reflection modeling | Retro-reflection modeling in 3DGS for accurate rendering of retro-reflective surfaces (signs, safety gear) |\r\n| Velox | arXiv'26 | 3D anisotropic | Velocity-aware 4D | Velocity-aware 4DGS for fast dynamic scene reconstruction with motion-adaptive temporal modeling |\r\n| 3DGS² | arXiv'26 | 3D anisotropic (super-sampled) | Super-sampling + temporal | Second-generation 3DGS with super-sampling and temporal interpolation for large-scale scenes |\r\n\r\n### Human & Avatar Methods\r\n\r\n| Method | Venue | Input | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| HumanSplatHMR | arXiv'26 | Image | Joint pose-avatar optimization; closes loop between HMR and differentiable rendering |\r\n| EmoTaG | CVPR'26 (2603.21332) | Image + audio | Emotion-aware talking head on GS |\r\n| SDTalk | arXiv'26 | Image + audio | Structured facial priors + dual-branch motion fields for Gaussian talking head |\r\n| HairGPT | SIGGRAPH'26 | Text/image | Strand-as-Language autoregressive modeling for 3D hairstyle synthesis |\r\n| D-Rex | SIGGRAPH'26 (2604.27871) | White-light avatar + target illumination | Decoupled relighting via LoRA fine-tuned video diffusion post-process; applicable to any white-light avatar system |\r\n\r\n### World Models & Spatial Intelligence\r\n\r\n_3DGS as world model primitive, differentiable simulation engine, or spatial intelligence representation_\r\n\r\nKey methods:\r\n- **GWM**: 3DGS as environment dynamics modeling primitive with autoregressive future state prediction\r\n- **FlashWorld**: Feed-forward 3DGS world model for real-time interactive 3D world generation\r\n- **GS-World**: 3DGS as differentiable simulation engine for world model + Sim2Real VLA\r\n- **Visionary**: WebGPU + 3DGS world model carrier platform for browser-native world model rendering\r\n- **RAD/DLWM**: 3DGS twin digital world for autonomous driving RL training\r\n\r\nComparison key: Does the method use 3DGS as (a) state representation only, (b) dynamics modeling primitive, or (c) differentiable simulation engine? This determines the depth of world model integration.\r\n\r\n### Autonomous Driving Methods\r\n\r\n| Method | Venue | Input | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| Real2Sim | arXiv'26 | 3D anisotropic (4D) | 4DGS + differentiable MPM | Physics-aware AD scene simulation with differentiable MPM for collision scenarios; bridges real-to-sim gap |\r\n| GaussianLSS | CVPR'25 | Multi-camera | GS for BEV perception |\r\n| Nighttime AD GS | ICRA'26 (2602.13549) | Nighttime multi-camera | PBR-based nighttime AD reconstruction |\r\n| ConFixGS | arXiv'26 (2605.09688) | Multi-camera | Confidence-aware diffusion for feedforward 3DGS fix; +3.68 dB PSNR on Waymo |\r\n\r\n### System & Infrastructure Methods\r\n\r\n| Method | Venue | Framework | Key Feature |\r\n|--------|-------|-----------|-------------|\r\n| VkSplat | Eurographics'26 | Vulkan | Vulkan-based 3DGS training; 3.3x speed; cross-vendor |\r\n| brush | Open-source | Rust/WebGPU/Burn | Cross-platform 3DGS training (Win/Mac/Linux/Android/Web); 4.3k stars; faster than gsplat |\r\n\r\n### Training Acceleration / Optimization Methods\r\n\r\n| Method | Venue | Strategy | Key Feature |\r\n|--------|-------|----------|-------------|\r\n| Structure-Aware Densification | SIGGRAPH'26 | Frequency-aware anisotropic splitting | Frequency-aware anisotropic splitting; multiview consistency; faster convergence |\r\n| GEMM-GS | DAC'26 (2604.02120) | Tensor Core GEMM | GPU acceleration via Tensor Cores; 1.42x speedup |\r\n| Denoising-GS | arXiv'26 (2605.14880) | Spatial-aware denoising | Spatial-aware denoising formulation for 3DGS optimization; spatial gradient + uncertainty-based pruning |\r\n| AdpSplit | arXiv'26 (2605.06876) | Error-driven adaptive split | Error-driven adaptive split operator; 9-22% training time reduction as drop-in replacement |\r\n\r\n### Real-Time NVS Methods\r\n\r\n| Method | Venue | Cameras | FPS | Latency | Key Feature |\r\n|--------|-------|---------|-----|---------|-------------|\r\n| 3DTV | arXiv'26 | 3 | 40 | 25ms | Delaunay-based triplet selection, real-time multi-camera synthesis |\r\n\r\n### Editing Methods\r\n\r\n| Method | Editing Type | Input | Quality |\r\n|--------|-------------|-------|---------|\r\n| GaussianEditor | Text/geometry | Mask + prompt | High |\r\n| GeoGaussian | Geometry | Mesh guidance | High |\r\n| Frosting | Appearance | Text prompt | Medium |\r\n| SketchFaceGS | Sketch-driven | 2D sketch | High (CVPR'26 Highlight) |\r\n| FluSplat | Text-driven | Sparse views | Medium-High |\r\n| TransSplat | Language-driven | Multi-view + text | High |\r\n| GOR-IS | Intrinsic-space removal | Image | High (+13% LPIPS) |\r\n| SVGS | arXiv'26 (2603.28126) | Text-driven 3D editing | Single view + text prompt | High |\r\n| VIRGi | TPAMI'26 (2603.02986) | Appearance editing | Image | View-dependent instant recoloring |\r\n| RDSplat | arXiv'26 (2512.06774) | Watermarking | Watermarked GS | Robust watermarking against diffusion editing |\r\n| FreeFix | arXiv'26 (2601.20857) | Diffusion guidance | No fine-tuning | Fine-tuning-free diffusion guidance for GS |\r\n\r\n## Output Format\r\n\r\nGenerate comparisons using this template:\r\n\r\n```\r\n## [Method A] vs [Method B] vs [Method C]\r\n\r\n### Overview Table\r\n| Dimension | Method A | Method B | Method C |\r\n|-----------|----------|----------|----------|\r\n| Primitive | ... | ... | ... |\r\n| Opacity | ... | ... | ... |\r\n| Rendering | ... | ... | ... |\r\n| ... | ... | ... | ... |\r\n\r\n### Detailed Analysis\r\n\r\n#### Primitive Representation\r\n[Paragraph comparing the fundamental representational differences]\r\n\r\n#### Design Trade-offs\r\n[Analysis of what each method gains and sacrifices]\r\n\r\n#### Recommendation\r\n- For novel view synthesis: [Best choice] because ...\r\n- For surface reconstruction: [Best choice] because ...\r\n- For real-time rendering: [Best choice] because ...\r\n```\r\n\r\n## Rules\r\n\r\n1. **Be technically precise**: Never oversimplify differences. If two methods differ in their opacity parameterization, explain exactly how.\r\n2. **Quote metrics when available**: Use actual numbers from papers, not estimates.\r\n3. **Avoid bias**: Present each method's strengths and weaknesses fairly.\r\n4. **Context matters**: A method that's worse on PSNR might be better for real-time. Always mention the use case.\r\n5. **Flag uncertainty**: If you don't have reliable data for a comparison dimension, say so explicitly.\r\n\r\n> If you like it, please star this repo https://github.com/jaccen/Awesome-Gaussian-Skills\n\nFile v1.4.7:_meta.json\n\n{\n  \"ownerId\": \"kn7509ejxr3dh88hw796a4hh6x85vvtc\",\n  \"slug\": \"3dgs-method-compare\",\n  \"version\": \"1.4.7\",\n  \"publishedAt\": 1779176237438\n}\n\nFile v1.4.7:skill-card.md\n\n## Description:\n\nCompare 3D Gaussian Splatting variants across 10+ dimensions with built-in knowledge of 523+ methods across 24 categories.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[jaccen](https://clawhub.ai/user/jaccen)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers, researchers, and technical evaluators use this skill to compare 3D Gaussian Splatting methods, summarize trade-offs, and choose methods for specific reconstruction, rendering, editing, robotics, or autonomous-driving scenarios.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The embedded 3DGS method database and paper metrics may become outdated or contain claims that need confirmation for important work.\n\nMitigation: Verify cited methods, paper claims, and numerical metrics against current primary sources before using the comparison for decisions.\n\nRisk: The artifact includes a promotional GitHub star request that may be inappropriate in some deployments.\n\nMitigation: Review public-facing skill text before deployment and remove promotional language if it conflicts with distribution policy.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/jaccen/skills/3dgs-method-compare)\n- [Awesome Gaussian Skills repository](https://github.com/jaccen/Awesome-Gaussian-Skills)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, guidance]\n\n**Output Format:** [Markdown comparison tables and narrative analysis]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May include recommendations, metric comparisons, and uncertainty notes when reliable data is unavailable.]\n\n## Skill Version(s):\n\n1.4.7 (source: frontmatter and server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment.\n\nArchive v1.4.4: 2 files, 8803 bytes\n\nFiles: SKILL.md (21696b), _meta.json (138b)\n\nFile v1.4.4:SKILL.md\n\n---\r\nname: 3dgs-method-compare\r\ndescription: \"Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 254+ methods across 21 categories.\"\r\nversion: 1.4.4\r\nauthor: jaccen\r\ntags: [\"3dgs\", \"gaussian-splatting\", \"method-comparison\", \"research\"]\r\n---\r\n\r\n# 3DGS Method Comparison Engine\r\n\r\nYou are an expert in 3D Gaussian Splatting methods with deep knowledge of 254+ variants. Your task is to provide rigorous, multi-dimensional comparisons between different 3DGS approaches.\r\n\r\n## Capabilities\r\n\r\n- Compare any combination of 3DGS variants across 10+ technical dimensions\r\n- Generate publication-quality comparison tables\r\n- Analyze design trade-offs and identify positioning\r\n- Provide recommendation based on specific use cases\r\n\r\n## Comparison Dimensions\r\n\r\nWhen comparing methods, analyze across the following dimensions:\r\n\r\n### 1. Primitive Representation\r\n- Shape: Full 3D Gaussian / 2D disk / 1D splat / hybrid / spatially-varying (SVGS)\r\n- Anisotropy: Isotropic / Anisotropic / Semi-anisotropic\r\n- Parameterization: (μ, Σ, opacity, SH) / (center, normal, scale, opacity) / custom / (μ, Σ, spatially-varying color+opacity, SH) (SVGS)\r\n\r\n### 2. Opacity / Alpha Mechanism\r\n- Range: [0, 1] / [-1, 1] / unbounded / sigmoid / tanh\r\n- Signed support: Yes (signed α) / No (standard GS)\r\n- Negative mechanism: Negative color (NegGS) / Negative opacity (signed) / None\r\n\r\n### 3. Color Representation\r\n- Spherical Harmonics order: 0/1/2/3\r\n- Color space: RGB / HDR / Feature vectors\r\n- Negative color support: Yes (NegGS) / No\r\n\r\n### 4. Rendering Formulation\r\n- Rasterization: Tile-based / Forward / Deferred\r\n- Blending: Front-to-back / Back-to-front\r\n- Anti-aliasing: EWA splatting / Mip-aware / None\r\n\r\n### 5. Frequency & Geometry Modeling\r\n- High-frequency boundary: Explicit / Implicit / None\r\n- Surface quality: Point-based / Surfels / Hybrid\r\n- Geometric constraints: Depth normal / ESDF / Mesh prior\r\n\r\n### 6. Density Control\r\n- Strategy: Clone + Split + Prune / Progressive / Anchor-based\r\n- Adaptivity: Gradient-based / Loss-based / Statistics-based\r\n- Compression: Pruning / Quantization / Distillation\r\n\r\n### 7. Training Strategy\r\n- Resolution schedule: Coarse-to-fine / Fixed\r\n- Iterations: 7k / 30k / custom\r\n- Regularization: Depth / Normal / Smoothness / Sparsity\r\n\r\n### 8. Performance Characteristics\r\n- Speed (FPS): Real-time (>30) / Interactive (10-30) / Offline (<10)\r\n- Memory: VRAM requirement\r\n- Storage: Model size (MB)\r\n- Scalability: Small object / Room-scale / City-scale\r\n\r\n### 9. Applicable Scenarios\r\n- Novel view synthesis\r\n- Surface reconstruction\r\n- 3D editing\r\n- Dynamic scenes\r\n- Large-scale scenes\r\n- Autonomous driving\r\n\r\n### 10. Code & Reproducibility\r\n- Official implementation available\r\n- Framework: PyTorch / JAX / CUDA / Custom\r\n- Dependencies\r\n\r\n## Known Methods Database\r\n\r\n### Foundation Methods\r\n\r\n| Method | Venue | Primitive | Opacity | Key Feature |\r\n|--------|-------|-----------|---------|-------------|\r\n| 3DGS | SIGGRAPH'23 | 3D anisotropic | [0,1] sigmoid | Tile-based rasterization |\r\n| Mip-Splatting | CVPR'24 (Best Student Paper) | 3D anisotropic + Mip | [0,1] | 3D smoothing + 2D Mip filter, alias-free |\r\n| 2DGS | SIGGRAPH'24 | 2D disk | [0,1] | Better surface reconstruction |\r\n| Scaffold-GS | ICCV'23 | Anchor+3D | [0,1] | Anchor-based scalability |\r\n| Scaffold-GS+ | CVPR'24 | Anchor+3D | [0,1] | Progressive training |\r\n| Softmax-GS | CVPR'26 (Findings) | 3D anisotropic | Softmax competition | Replaces α-compositing with learnable softmax; blend-vs-bound |\r\n| LeGS | arXiv'26 | 3D anisotropic | RL-controlled | RL-based learnable density control replacing heuristics; O(N) reward |\r\n\r\n### Signed / Decomposed Methods\r\n\r\n| Method | Opacity Range | Color Range | Mechanism |\r\n|--------|--------------|-------------|-----------|\r\n| NegGS | [0, +∞) (non-negative) | ℝ (negative allowed) | Negative color + Diff-Gaussian |\r\n| (Standard GS) | [0, 1] via sigmoid | [0, +∞) | Standard α-compositing |\r\n\r\n**Critical Distinction**: Methods using \"negative\" concepts differ fundamentally:\r\n- **Signed opacity (α ∈ [-1,1])**: Opacity α can be negative, rendering formula modified. The Gaussian primitive itself carries a sign. Better for sharp geometric boundaries.\r\n- **NegGS**: Opacity remains non-negative, but color values can be negative. Uses Diff-Gaussian (subtraction of two Gaussians) to model ring/crescent structures.\r\n\r\n### Compression Methods\r\n\r\n| Method | Compression Ratio | Quality Impact | Speed |\r\n|--------|-------------------|----------------|-------|\r\n| Compact-3DGS | 10-15x | Minimal PSNR drop | Faster |\r\n| LightGS | 15-20x | Slight drop | Much faster |\r\n| MobileGS | 50-100x | Moderate drop | Real-time mobile |\r\n| Embedded-3DGS | 10x | Minimal | Comparable |\r\n| HAC | ~100x | Slight drop | Faster after decode |\r\n| OT-UVGS | UV tensor | ↑ vs spherical UVGS | Same as UVGS |\r\n| NanoGS | Training-free | Minimal (KNN merge) | CPU-only, instant |\r\n| MesonGS++ | 34x | Minimal | Faster after decode (0-1 ILP hyperparameter search) |\r\n| GETA-3DGS | 5x | Minimal | First end-to-end automatic joint structured pruning + quantization; QADG; render-aware saliency |\r\n| CAGS | ~7x (streaming) | Minimal | VQ-based compression with Level-of-Detail streaming; progressive decode for bandwidth-adaptive deployment |\r\n\r\n### Robustness / Regularization Methods\r\n\r\n| Method | Venue | Prior Source | Key Feature |\r\n|--------|-------|-------------|-------------|\r\n| EnerGS | arXiv'26 | LiDAR (partial geometric) | Energy-based soft guidance instead of hard constraints; improves outdoor large-scale scenes |\r\n| Luminance-GS++ | TPAMI'26 | Illumination prior | Illumination-robust NVS; decouples shading from geometry |\r\n\r\n### Geometry / Surface Methods\r\n\r\n| Method | Venue | Surface Quality | Key Feature |\r\n|--------|-------|----------------|-------------|\r\n| 2DGS | SIGGRAPH'24 | High | Oriented 2D disks for geometry |\r\n| SuGaR | CVPR'24 | High | Surface-aligned regularization |\r\n| PGSR | TVCG'24 | Highest (SOTA) | Planar regularizer + unbiased depth rendering |\r\n| PAGaS | arXiv'26 | High (depth) | 1DoF Gaussians for depth refinement |\r\n| Vol3DGS | CVPR'25 | High | Volume-consistent rendering |\r\n| 2D-SuGaR | arXiv'26 | Highest (DTU SOTA) | 2DGS + monocular depth/normal priors; depth-guided init; clustering-based pruning |\r\n| IRIS | arXiv'26 (2603.15368) | Hybrid | GS-proxy neural field with analytical ray intersection; hybrid rendering |\r\n| DiffSoup | arXiv'26 (2603.27151) | Extreme simplification | Triangle soup as alternative primitive to Gaussians |\r\n| 3DSS | arXiv'26 (2605.05876) | High (inverse rendering) | First differentiable surface splatting; coverage-based compositing from EWA; joint shape+SVBRDF+lighting |\r\n| SVGS | arXiv'24 (2411.18966) | High (Blender SOTA) | Spatially varying color+opacity within each Gaussian; movable kernels (1.4x params); >30 FPS |\r\n| AmbiSuR | ICML'26 | High (photometric) | Photometric ambiguity disambiguation for accurate GS surface reconstruction |\r\n| DySurface | arXiv'26 | High (4D surface) | Bridges explicit Gaussians and implicit SDF for consistent 4D surface reconstruction |\r\n\r\n### Generation / Text-to-3D\r\n\r\n| Method | Venue | Input | Output | Key Feature |\r\n|--------|-------|-------|--------|-------------|\r\n| DreamGaussian | ICLR'24 (Oral) | Text prompt | 3D mesh + 3DGS | SDS + 3DGS prior, seconds |\r\n| GaussianEditor | Preprint | Text/geometry mask | Edited 3DGS | CLIP-guided selection + editing |\r\n| ArtifactWorld | arXiv'26 (2604.12251) | Artifact images | Restored video | Video generation for artifact restoration |\r\n| SceneGen-LLMRL | arXiv'26 (2605.05711) | Language | Interactive 3D scene | LLM-RL coupling for unified 3D scene generation + immersive interaction |\r\n\r\n### Language / Semantic\r\n\r\n| Method | Venue | Feature Source | 3D Storage | Key Feature |\r\n|--------|-------|---------------|------------|-------------|\r\n| LangSplat | CVPR'24 | CLIP (2D distillation) | Per-Gaussian CLIP features | Open-vocabulary 3D queries |\r\n| Feature 3DGS | CVPR'24 | DINO/SAM (2D distillation) | Per-Gaussian feature vectors | Downstream task features |\r\n| NRGS | arXiv'26 | Neural network | Learned regularization | Robust semantic 3DGS |\r\n| Semantic Foam | CVPR'26 (Highlight) | Volumetric Voronoi mesh | Per-cell semantic feature field | Semantic decomposition; outperforms Gaussian Grouping, SAGA |\r\n| GLMap | CVPR'26 | Multi-scale semantics | Per-Gaussian language features | Gaussian-Language Map; zero-shot navigation |\r\n| NG-GS | arXiv'26 (2604.14706) | NeRF-guided | Per-Gaussian segmentation | NeRF-guided GS segmentation |\r\n| PointGS | CVPR'26 | SAM masks (contrastive distillation) | Per-Gaussian semantic features | 3DGS as unified intermediate for unsupervised 3D point cloud segmentation; SAM→3D contrastive learning |\r\n\r\n### Feed-Forward Methods\r\n\r\n| Method | Venue | #Gaussians | Inference | Key Feature |\r\n|--------|-------|------------|-----------|-------------|\r\n| GlobalSplat | Preprint'26 | ~16K | <78ms | Global scene tokens, 4MB footprint |\r\n| MVSplat | ECCV'24 | Variable | Single-pass | Cost-volume-based prediction |\r\n| GS-LRM | ECCV'24 | Variable | Single-pass | 1B transformer, zero-shot generalization |\r\n| DepthSplat | CVPR'25 | Variable | Single-pass | Stereo-guided depth regularization |\r\n| InstantSplat | arXiv'24 | Variable | ~40s total | Pose-free sparse-view |\r\n| AnySplat | SIGGRAPH'25 | Variable | Single-pass | In-the-wild unconstrained views |\r\n| SparseSplat | CVPR'26 | 22% of SOTA | Single-pass | Pixel-unaligned, entropy-based probabilistic sampling, 3D-Local Attribute Predictor |\r\n| OT-UVGS | EG'26 | UV tensor | Same as UVGS | OT-based UV mapping, O(N log N) |\r\n| Free Geometry | arXiv'26 | Adaptive | Single-pass + LoRA | Self-evolving feed-forward, +3.73% camera accuracy |\r\n| FTSplat | arXiv'26 (2603.05932) | Variable | Single-pass | Feed-forward triangle splatting |\r\n\r\n### SLAM Methods\r\n\r\n| Method | Venue | Input | Scale | Key Feature |\r\n|--------|-------|-------|-------|-------------|\r\n| Gaussian Splatting SLAM | CVPR'24 (Highlight) | Monocular video | Room-scale | First real-time monocular 3DGS SLAM, differentiable rendering for joint pose+map |\r\n| CGS-SLAM | IROS'25 | Monocular video | Room-scale | Voxel-based compact representation for efficiency |\r\n| WildGS-SLAM | CVPR'25 | Monocular video | Room-scale | Dynamic environments, uncertainty-aware mapping via pretrained 3D priors |\r\n| S3PO-GS | ICCV'25 | Monocular video | Outdoor | Scale-consistent pose optimization, eliminates outdoor scale drift |\r\n| Flow4DGS-SLAM | arXiv'26 | Monocular video | Room-scale | Optical flow-guided 4DGS for temporal consistency |\r\n| E2EGS | CVPR'26 (2603.14684) | Event camera | Room-scale | Event-camera pose-free 3D reconstruction |\r\n| MAGS-SLAM | arXiv'26 | RGB (multi-agent) | Multi-room | First RGB-only multi-agent 3DGS SLAM; compact submap communication + geometry/appearance-aware loop verification |\r\n\r\n### Large-Scale Methods\r\n\r\n| Method | Venue | Scale | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| Scaffold-GS | ICCV'23 | Building | Anchor-based efficiency |\r\n| Scaffold-GS+ | CVPR'24 | City | Progressive training |\r\n| CityGaussian | ECCV'24 | City | Hierarchical LOD |\r\n| Street Gaussians | ECCV'24 | Street | Static/dynamic decomposition, driving scenes |\r\n| Octree-GS | Preprint | City | Octree acceleration + LOD |\r\n\r\n### Cross-Domain Applications\r\n\r\n| Method | Venue | Domain | Key Feature |\r\n|--------|-------|--------|-------------|\r\n| GS-DOT | arXiv'26 | Medical (DOT) | Diffusion transport for photon imaging |\r\n| BiSplat-WRF | IEEE ICC'26 Workshop | Wireless (WRF) | Planar GS + bilinear spatial transformer for EM coupling |\r\n| FieryGS | ICLR'26 | Physics simulation | Physics-integrated fire synthesis |\r\n| SplAttN | ICML'26 (Spotlight) | Point cloud completion | Gaussian soft splatting for point cloud completion |\r\n| Fake3DGS | ICPR'26 | Forensics | First benchmark for 3D manipulation detection in neural rendering |\r\n| SandSim | arXiv'26 | Digital art | Curve-guided Gaussian for sand painting reconstruction |\r\n| RGS | arXiv'26 | Medical (CBCT) | Residual wavelet-GS for sparse-view CBCT |\r\n| RESPIRE | arXiv'26 | Medical (bronchoscopy) | CT-informed mesh-anchored GS for dynamic bronchoscopy |\r\n| Color-Encoded Illumination | CVPR'26 (Highlight) | High-speed imaging | Color-coded temporal info for volumetric reconstruction |\r\n| HDR-NSFF | ICLR'26 (2603.08313) | Dynamic HDR scenes | HDR dynamic scene neural scene flow fields |\r\n| 3DGS AD Safety Eval | SafeComp'26 | Autonomous driving | Industrial fidelity evaluation for AD perception |\r\n| HeroGS | CVPR'26 | Sparse-view NVS | Hierarchical guidance for sparse-view robust 3DGS |\r\n| Sparse-View 3DGS Wild | arXiv'26 | Sparse-view NVS | Diffusion-guided sparse-view enhancement |\r\n| Pi-GS | arXiv'26 (2602.03327) | Sparse-view NVS | Sparse-view with π³ reference-free initialization |\r\n| GS-Surrogate | arXiv'26 (2604.06358) | Physics simulation | Deformable GS for simulation visualization |\r\n| 3DGEER | ICLR'26 | Rendering (exact) | Exact ray-Gaussian rendering replacing splatting; fisheye/generic camera support; top 1% |\r\n| Forecast-GS | arXiv'26 | Robotics | Predictive GS for forecasting task-completed states in robotic manipulation |\r\n| GeoQuery | SIGGRAPH'26 | Sparse-view NVS | Geometry-guided cross-view attention with geometry-aligned proxy queries from predicted depth |\r\n| PairDropGS | arXiv'26 | Sparse-view NVS | Paired dropout-induced consistency regularization with progressive scheduling |\r\n| VidSplat | SIGGRAPH'26 | Sparse-view NVS | Training-free generative framework leveraging video diffusion priors with iterative confidence refinement |\r\n\r\n### Dynamic / 4DGS Methods\r\n\r\n| Method | Venue | Primitive | Rendering | Key Feature |\r\n|--------|-------|-----------|-----------|-------------|\r\n| FreeTimeGS++ | arXiv'26 (2605.03337) | 4D Gaussians + durations | Gated marginalization | Neural velocity fields + emergent temporal partitioning; comprehensive 4DGS analysis |\r\n| ParticleGS | arXiv'26 | 3D anisotropic + physics | Standard α-compositing | Physics-based motion extrapolation for fluid/dynamic scenes; Lagrangian particle dynamics |\r\n| TransmissiveGS | arXiv'26 | Dual-GS (surface + reflection) | Deferred shading | Transmissive + reflective dual decomposition; separate G-buffer compositing for glass/refractive objects |\r\n| PD-4DGS | arXiv'26 | 3-layer progressive (static + global deform + local refine) | Progressive streaming | DASH/HLS-compatible 4DGS streaming; ~1.7s first-frame latency vs 73-930s monolithic |\r\n| 3DGS³ | arXiv'26 | 3D anisotropic (super-sampled) | Standard + temporal interpolation | Gradient-Aware Super Sampling + Lightweight Temporal Frame Interpolation for large-scale 3DGS |\r\n| BlitzGS | arXiv'26 | 3D anisotropic (distributed) | Parity-based multi-GPU | Distributed city-scale GS training; parity-based sharding across multi-GPU; eliminates single-GPU memory bottleneck |\r\n| Z-Order GS | arXiv'26 | 3D anisotropic (Z-ordered) | Z-order curve indexing | Z-order curve spatial indexing for cache-coherent Gaussian traversal; improved rendering throughput |\r\n| PanoPlane | arXiv'26 | Planar (panoramic) | Plane-based compositing | Panoramic plane-based GS for omnidirectional NVS; efficient panoramic scene representation |\r\n| SparseOIT | arXiv'26 | 3D anisotropic | Order-independent transparency | Sparse order-independent transparency for correct See-through rendering of overlapping semi-transparent Gaussians |\r\n| SCOUP | arXiv'26 | Sparse code primitives | Language-conditioned | Sparse code language GS; language-conditioned sparse coding for controllable 3DGS generation |\r\n| AV1-3DGS | arXiv'26 | 3D anisotropic | AV1 motion-vector SfM | AV1 codec motion vectors for dense SfM; 63% training time reduction; leverages video compression priors |\r\n| RoSplat | arXiv'26 | 3D anisotropic (feed-forward) | Pixel-wise GS | Feed-forward pixel-wise GS for sparse-view NVS; requires alpha normalization for varying view counts |\r\n| HarmoGS | arXiv'26 | 3D anisotropic | Harmonized optimization | Gradient harmonization for in-the-wild 3DGS; resolves cross-view gradient conflicts from transient distractors and illumination inconsistencies |\r\n| GuardMarkGS | arXiv'26 | 3D anisotropic | Watermark + deterrence | First unified watermarking + edit deterrence framework for 3DGS assets; security for 3D content |\r\n| FaceParts | arXiv'26 | 3D anisotropic (part-based) | Part-compositional | Part-based decomposable Gaussian avatar; modular facial region modeling for expressive avatars |\r\n| RetroNVS | arXiv'26 | 3D anisotropic | Retro-reflection modeling | Retro-reflection modeling in 3DGS for accurate rendering of retro-reflective surfaces (signs, safety gear) |\r\n| Velox | arXiv'26 | 3D anisotropic | Velocity-aware 4D | Velocity-aware 4DGS for fast dynamic scene reconstruction with motion-adaptive temporal modeling |\r\n| 3DGS² | arXiv'26 | 3D anisotropic (super-sampled) | Super-sampling + temporal | Second-generation 3DGS with super-sampling and temporal interpolation for large-scale scenes |\r\n\r\n### Human & Avatar Methods\r\n\r\n| Method | Venue | Input | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| HumanSplatHMR | arXiv'26 | Image | Joint pose-avatar optimization; closes loop between HMR and differentiable rendering |\r\n| EmoTaG | CVPR'26 (2603.21332) | Image + audio | Emotion-aware talking head on GS |\r\n| SDTalk | arXiv'26 | Image + audio | Structured facial priors + dual-branch motion fields for Gaussian talking head |\r\n| HairGPT | SIGGRAPH'26 | Text/image | Strand-as-Language autoregressive modeling for 3D hairstyle synthesis |\r\n| D-Rex | SIGGRAPH'26 (2604.27871) | White-light avatar + target illumination | Decoupled relighting via LoRA fine-tuned video diffusion post-process; applicable to any white-light avatar system |\r\n\r\n### Autonomous Driving Methods\r\n\r\n| Method | Venue | Input | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| Real2Sim | arXiv'26 | 3D anisotropic (4D) | 4DGS + differentiable MPM | Physics-aware AD scene simulation with differentiable MPM for collision scenarios; bridges real-to-sim gap |\r\n| GaussianLSS | CVPR'25 | Multi-camera | GS for BEV perception |\r\n| Nighttime AD GS | ICRA'26 (2602.13549) | Nighttime multi-camera | PBR-based nighttime AD reconstruction |\r\n\r\n### System & Infrastructure Methods\r\n\r\n| Method | Venue | Framework | Key Feature |\r\n|--------|-------|-----------|-------------|\r\n| VkSplat | Eurographics'26 | Vulkan | Vulkan-based 3DGS training; 3.3x speed; cross-vendor |\r\n| brush | Open-source | Rust/WebGPU/Burn | Cross-platform 3DGS training (Win/Mac/Linux/Android/Web); 4.3k stars; faster than gsplat |\r\n\r\n### Training Acceleration Methods\r\n\r\n| Method | Venue | Strategy | Key Feature |\r\n|--------|-------|----------|-------------|\r\n| Structure-Aware Densification | SIGGRAPH'26 | Frequency-aware anisotropic splitting | Frequency-aware anisotropic splitting; multiview consistency; faster convergence |\r\n| GEMM-GS | DAC'26 (2604.02120) | Tensor Core GEMM | GPU acceleration via Tensor Cores; 1.42x speedup |\r\n\r\n### Real-Time NVS Methods\r\n\r\n| Method | Venue | Cameras | FPS | Latency | Key Feature |\r\n|--------|-------|---------|-----|---------|-------------|\r\n| 3DTV | arXiv'26 | 3 | 40 | 25ms | Delaunay-based triplet selection, real-time multi-camera synthesis |\r\n\r\n### Editing Methods\r\n\r\n| Method | Editing Type | Input | Quality |\r\n|--------|-------------|-------|---------|\r\n| GaussianEditor | Text/geometry | Mask + prompt | High |\r\n| GeoGaussian | Geometry | Mesh guidance | High |\r\n| Frosting | Appearance | Text prompt | Medium |\r\n| SketchFaceGS | Sketch-driven | 2D sketch | High (CVPR'26 Highlight) |\r\n| FluSplat | Text-driven | Sparse views | Medium-High |\r\n| TransSplat | Language-driven | Multi-view + text | High |\r\n| GOR-IS | Intrinsic-space removal | Image | High (+13% LPIPS) |\r\n| SVGS | arXiv'26 (2603.28126) | Text-driven 3D editing | Single view + text prompt | High |\r\n| VIRGi | TPAMI'26 (2603.02986) | Appearance editing | Image | View-dependent instant recoloring |\r\n| RDSplat | arXiv'26 (2512.06774) | Watermarking | Watermarked GS | Robust watermarking against diffusion editing |\r\n| FreeFix | arXiv'26 (2601.20857) | Diffusion guidance | No fine-tuning | Fine-tuning-free diffusion guidance for GS |\r\n\r\n## Output Format\r\n\r\nGenerate comparisons using this template:\r\n\r\n```\r\n## [Method A] vs [Method B] vs [Method C]\r\n\r\n### Overview Table\r\n| Dimension | Method A | Method B | Method C |\r\n|-----------|----------|----------|----------|\r\n| Primitive | ... | ... | ... |\r\n| Opacity | ... | ... | ... |\r\n| Rendering | ... | ... | ... |\r\n| ... | ... | ... | ... |\r\n\r\n### Detailed Analysis\r\n\r\n#### Primitive Representation\r\n[Paragraph comparing the fundamental representational differences]\r\n\r\n#### Design Trade-offs\r\n[Analysis of what each method gains and sacrifices]\r\n\r\n#### Recommendation\r\n- For novel view synthesis: [Best choice] because ...\r\n- For surface reconstruction: [Best choice] because ...\r\n- For real-time rendering: [Best choice] because ...\r\n```\r\n\r\n## Rules\r\n\r\n1. **Be technically precise**: Never oversimplify differences. If two methods differ in their opacity parameterization, explain exactly how.\r\n2. **Quote metrics when available**: Use actual numbers from papers, not estimates.\r\n3. **Avoid bias**: Present each method's strengths and weaknesses fairly.\r\n4. **Context matters**: A method that's worse on PSNR might be better for real-time. Always mention the use case.\r\n5. **Flag uncertainty**: If you don't have reliable data for a comparison dimension, say so explicitly.\r\n\r\n> If you like it, please star this repo https://github.com/jaccen/Awesome-Gaussian-Skills\n\nFile v1.4.4:_meta.json\n\n{\n  \"ownerId\": \"kn7509ejxr3dh88hw796a4hh6x85vvtc\",\n  \"slug\": \"3dgs-method-compare\",\n  \"version\": \"1.4.4\",\n  \"publishedAt\": 1778929179116\n}\n\nArchive v0.1.2: 2 files, 6769 bytes\n\nFiles: SKILL.md (16007b), _meta.json (138b)\n\nFile v0.1.2:SKILL.md\n\n---\r\nname: 3dgs-method-compare\r\ndescription: Compare 3D Gaussian Splatting variants across multiple dimensions. Generates detailed comparison tables covering primitive representation, rendering formulation, training strategy, and performance. Built-in knowledge of 150+ 3DGS methods.\r\nversion: 1.3.0\r\nauthor: jaccen\r\ntags:\r\n  - 3dgs\r\n  - gaussian-splatting\r\n  - method-comparison\r\n  - research\r\n  - nerf\r\ntrigger:\r\n  - \"对比\"\r\n  - \"比较\"\r\n  - \"compare\"\r\n  - \"difference between\"\r\n  - \"和...有什么区别\"\r\n  - \"哪个方法更好\"\r\n  - \"method comparison\"\r\n  - \"GS vs\"\r\n  - \"3DGS vs 2DGS\"\r\n---\r\n\r\n# 3DGS Method Comparison Engine\r\n\r\nYou are an expert in 3D Gaussian Splatting methods with deep knowledge of 150+ variants. Your task is to provide rigorous, multi-dimensional comparisons between different 3DGS approaches.\r\n\r\n## Capabilities\r\n\r\n- Compare any combination of 3DGS variants across 10+ technical dimensions\r\n- Generate publication-quality comparison tables\r\n- Analyze design trade-offs and identify positioning\r\n- Provide recommendation based on specific use cases\r\n\r\n## Comparison Dimensions\r\n\r\nWhen comparing methods, analyze across the following dimensions:\r\n\r\n### 1. Primitive Representation\r\n- Shape: Full 3D Gaussian / 2D disk / 1D splat / hybrid\r\n- Anisotropy: Isotropic / Anisotropic / Semi-anisotropic\r\n- Parameterization: (μ, Σ, opacity, SH) / (center, normal, scale, opacity) / custom\r\n\r\n### 2. Opacity / Alpha Mechanism\r\n- Range: [0, 1] / [-1, 1] / unbounded / sigmoid / tanh\r\n- Signed support: Yes (signed α) / No (standard GS)\r\n- Negative mechanism: Negative color (NegGS) / Negative opacity (signed) / None\r\n\r\n### 3. Color Representation\r\n- Spherical Harmonics order: 0/1/2/3\r\n- Color space: RGB / HDR / Feature vectors\r\n- Negative color support: Yes (NegGS) / No\r\n\r\n### 4. Rendering Formulation\r\n- Rasterization: Tile-based / Forward / Deferred\r\n- Blending: Front-to-back / Back-to-front\r\n- Anti-aliasing: EWA splatting / Mip-aware / None\r\n\r\n### 5. Frequency & Geometry Modeling\r\n- High-frequency boundary: Explicit / Implicit / None\r\n- Surface quality: Point-based / Surfels / Hybrid\r\n- Geometric constraints: Depth normal / ESDF / Mesh prior\r\n\r\n### 6. Density Control\r\n- Strategy: Clone + Split + Prune / Progressive / Anchor-based\r\n- Adaptivity: Gradient-based / Loss-based / Statistics-based\r\n- Compression: Pruning / Quantization / Distillation\r\n\r\n### 7. Training Strategy\r\n- Resolution schedule: Coarse-to-fine / Fixed\r\n- Iterations: 7k / 30k / custom\r\n- Regularization: Depth / Normal / Smoothness / Sparsity\r\n\r\n### 8. Performance Characteristics\r\n- Speed (FPS): Real-time (>30) / Interactive (10-30) / Offline (<10)\r\n- Memory: VRAM requirement\r\n- Storage: Model size (MB)\r\n- Scalability: Small object / Room-scale / City-scale\r\n\r\n### 9. Applicable Scenarios\r\n- Novel view synthesis\r\n- Surface reconstruction\r\n- 3D editing\r\n- Dynamic scenes\r\n- Large-scale scenes\r\n- Autonomous driving\r\n\r\n### 10. Code & Reproducibility\r\n- Official implementation available\r\n- Framework: PyTorch / JAX / CUDA / Custom\r\n- Dependencies\r\n\r\n## Known Methods Database\r\n\r\n### Foundation Methods\r\n\r\n| Method | Venue | Primitive | Opacity | Key Feature |\r\n|--------|-------|-----------|---------|-------------|\r\n| 3DGS | SIGGRAPH'23 | 3D anisotropic | [0,1] sigmoid | Tile-based rasterization |\r\n| Mip-Splatting | CVPR'24 (Best Student Paper) | 3D anisotropic + Mip | [0,1] | 3D smoothing + 2D Mip filter, alias-free |\r\n| 2DGS | SIGGRAPH'24 | 2D disk | [0,1] | Better surface reconstruction |\r\n| Scaffold-GS | ICCV'23 | Anchor+3D | [0,1] | Anchor-based scalability |\r\n| Scaffold-GS+ | CVPR'24 | Anchor+3D | [0,1] | Progressive training |\r\n| Softmax-GS | CVPR'26 (Findings) | 3D anisotropic | Softmax competition | Replaces α-compositing with learnable softmax; blend-vs-bound |\r\n| LeGS | arXiv'26 | 3D anisotropic | RL-controlled | RL-based learnable density control replacing heuristics; O(N) reward |\r\n\r\n### Signed / Decomposed Methods\r\n\r\n| Method | Opacity Range | Color Range | Mechanism |\r\n|--------|--------------|-------------|-----------|\r\n| NegGS | [0, +∞) (non-negative) | ℝ (negative allowed) | Negative color + Diff-Gaussian |\r\n| (Standard GS) | [0, 1] via sigmoid | [0, +∞) | Standard α-compositing |\r\n\r\n**Critical Distinction**: Methods using \"negative\" concepts differ fundamentally:\r\n- **Signed opacity (α ∈ [-1,1])**: Opacity α can be negative, rendering formula modified. The Gaussian primitive itself carries a sign. Better for sharp geometric boundaries.\r\n- **NegGS**: Opacity remains non-negative, but color values can be negative. Uses Diff-Gaussian (subtraction of two Gaussians) to model ring/crescent structures.\r\n\r\n### Compression Methods\r\n\r\n| Method | Compression Ratio | Quality Impact | Speed |\r\n|--------|-------------------|----------------|-------|\r\n| Compact-3DGS | 10-15x | Minimal PSNR drop | Faster |\r\n| LightGS | 15-20x | Slight drop | Much faster |\r\n| MobileGS | 50-100x | Moderate drop | Real-time mobile |\r\n| Embedded-3DGS | 10x | Minimal | Comparable |\r\n| HAC | ~100x | Slight drop | Faster after decode |\r\n| OT-UVGS | UV tensor | ↑ vs spherical UVGS | Same as UVGS |\r\n| NanoGS | Training-free | Minimal (KNN merge) | CPU-only, instant |\r\n| MesonGS++ | 34x | Minimal | Faster after decode (0-1 ILP hyperparameter search) |\r\n| GETA-3DGS | 5x | Minimal | First end-to-end automatic joint structured pruning + quantization; QADG; render-aware saliency |\r\n\r\n### Robustness / Regularization Methods\r\n\r\n| Method | Venue | Prior Source | Key Feature |\r\n|--------|-------|-------------|-------------|\r\n| EnerGS | arXiv'26 | LiDAR (partial geometric) | Energy-based soft guidance instead of hard constraints; improves outdoor large-scale scenes |\r\n| Luminance-GS++ | TPAMI'26 | Illumination prior | Illumination-robust NVS; decouples shading from geometry |\r\n\r\n### Geometry / Surface Methods\r\n\r\n| Method | Venue | Surface Quality | Key Feature |\r\n|--------|-------|----------------|-------------|\r\n| 2DGS | SIGGRAPH'24 | High | Oriented 2D disks for geometry |\r\n| SuGaR | CVPR'24 | High | Surface-aligned regularization |\r\n| PGSR | TVCG'24 | Highest (SOTA) | Planar regularizer + unbiased depth rendering |\r\n| PAGaS | arXiv'26 | High (depth) | 1DoF Gaussians for depth refinement |\r\n| Vol3DGS | CVPR'25 | High | Volume-consistent rendering |\r\n| 2D-SuGaR | arXiv'26 | Highest (DTU SOTA) | 2DGS + monocular depth/normal priors; depth-guided init; clustering-based pruning |\r\n| IRIS | arXiv'26 (2603.15368) | Hybrid | GS-proxy neural field with analytical ray intersection; hybrid rendering |\r\n| DiffSoup | arXiv'26 (2603.27151) | Extreme simplification | Triangle soup as alternative primitive to Gaussians |\r\n\r\n### Generation / Text-to-3D\r\n\r\n| Method | Venue | Input | Output | Key Feature |\r\n|--------|-------|-------|--------|-------------|\r\n| DreamGaussian | ICLR'24 (Oral) | Text prompt | 3D mesh + 3DGS | SDS + 3DGS prior, seconds |\r\n| GaussianEditor | Preprint | Text/geometry mask | Edited 3DGS | CLIP-guided selection + editing |\r\n| ArtifactWorld | arXiv'26 (2604.12251) | Artifact images | Restored video | Video generation for artifact restoration |\r\n\r\n### Language / Semantic\r\n\r\n| Method | Venue | Feature Source | 3D Storage | Key Feature |\r\n|--------|-------|---------------|------------|-------------|\r\n| LangSplat | CVPR'24 | CLIP (2D distillation) | Per-Gaussian CLIP features | Open-vocabulary 3D queries |\r\n| Feature 3DGS | CVPR'24 | DINO/SAM (2D distillation) | Per-Gaussian feature vectors | Downstream task features |\r\n| NRGS | arXiv'26 | Neural network | Learned regularization | Robust semantic 3DGS |\r\n| Semantic Foam | CVPR'26 (Highlight) | Volumetric Voronoi mesh | Per-cell semantic feature field | Semantic decomposition; outperforms Gaussian Grouping, SAGA |\r\n| GLMap | CVPR'26 | Multi-scale semantics | Per-Gaussian language features | Gaussian-Language Map; zero-shot navigation |\r\n| NG-GS | arXiv'26 (2604.14706) | NeRF-guided | Per-Gaussian segmentation | NeRF-guided GS segmentation |\r\n\r\n### Feed-Forward Methods\r\n\r\n| Method | Venue | #Gaussians | Inference | Key Feature |\r\n|--------|-------|------------|-----------|-------------|\r\n| GlobalSplat | Preprint'26 | ~16K | <78ms | Global scene tokens, 4MB footprint |\r\n| MVSplat | ECCV'24 | Variable | Single-pass | Cost-volume-based prediction |\r\n| GS-LRM | ECCV'24 | Variable | Single-pass | 1B transformer, zero-shot generalization |\r\n| DepthSplat | CVPR'25 | Variable | Single-pass | Stereo-guided depth regularization |\r\n| InstantSplat | arXiv'24 | Variable | ~40s total | Pose-free sparse-view |\r\n| AnySplat | SIGGRAPH'25 | Variable | Single-pass | In-the-wild unconstrained views |\r\n| SparseSplat | CVPR'26 | 22% of SOTA | Single-pass | Pixel-unaligned, entropy-based probabilistic sampling, 3D-Local Attribute Predictor |\r\n| OT-UVGS | EG'26 | UV tensor | Same as UVGS | OT-based UV mapping, O(N log N) |\r\n| Free Geometry | arXiv'26 | Adaptive | Single-pass + LoRA | Self-evolving feed-forward, +3.73% camera accuracy |\r\n| FTSplat | arXiv'26 (2603.05932) | Variable | Single-pass | Feed-forward triangle splatting |\r\n\r\n### SLAM Methods\r\n\r\n| Method | Venue | Input | Scale | Key Feature |\r\n|--------|-------|-------|-------|-------------|\r\n| Gaussian Splatting SLAM | CVPR'24 (Highlight) | Monocular video | Room-scale | First real-time monocular 3DGS SLAM, differentiable rendering for joint pose+map |\r\n| CGS-SLAM | IROS'25 | Monocular video | Room-scale | Voxel-based compact representation for efficiency |\r\n| WildGS-SLAM | CVPR'25 | Monocular video | Room-scale | Dynamic environments, uncertainty-aware mapping via pretrained 3D priors |\r\n| S3PO-GS | ICCV'25 | Monocular video | Outdoor | Scale-consistent pose optimization, eliminates outdoor scale drift |\r\n| Flow4DGS-SLAM | arXiv'26 | Monocular video | Room-scale | Optical flow-guided 4DGS for temporal consistency |\r\n| E2EGS | CVPR'26 (2603.14684) | Event camera | Room-scale | Event-camera pose-free 3D reconstruction |\r\n\r\n### Large-Scale Methods\r\n\r\n| Method | Venue | Scale | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| Scaffold-GS | ICCV'23 | Building | Anchor-based efficiency |\r\n| Scaffold-GS+ | CVPR'24 | City | Progressive training |\r\n| CityGaussian | ECCV'24 | City | Hierarchical LOD |\r\n| Street Gaussians | ECCV'24 | Street | Static/dynamic decomposition, driving scenes |\r\n| Octree-GS | Preprint | City | Octree acceleration + LOD |\r\n\r\n### Cross-Domain Applications\r\n\r\n| Method | Venue | Domain | Key Feature |\r\n|--------|-------|--------|-------------|\r\n| GS-DOT | arXiv'26 | Medical (DOT) | Diffusion transport for photon imaging |\r\n| BiSplat-WRF | IEEE ICC'26 Workshop | Wireless (WRF) | Planar GS + bilinear spatial transformer for EM coupling |\r\n| FieryGS | ICLR'26 | Physics simulation | Physics-integrated fire synthesis |\r\n| SplAttN | ICML'26 (Spotlight) | Point cloud completion | Gaussian soft splatting for point cloud completion |\r\n| Fake3DGS | ICPR'26 | Forensics | First benchmark for 3D manipulation detection in neural rendering |\r\n| SandSim | arXiv'26 | Digital art | Curve-guided Gaussian for sand painting reconstruction |\r\n| RGS | arXiv'26 | Medical (CBCT) | Residual wavelet-GS for sparse-view CBCT |\r\n| RESPIRE | arXiv'26 | Medical (bronchoscopy) | CT-informed mesh-anchored GS for dynamic bronchoscopy |\r\n| Color-Encoded Illumination | CVPR'26 (Highlight) | High-speed imaging | Color-coded temporal info for volumetric reconstruction |\r\n| HDR-NSFF | ICLR'26 (2603.08313) | Dynamic HDR scenes | HDR dynamic scene neural scene flow fields |\r\n| 3DGS AD Safety Eval | SafeComp'26 | Autonomous driving | Industrial fidelity evaluation for AD perception |\r\n| HeroGS | CVPR'26 | Sparse-view NVS | Hierarchical guidance for sparse-view robust 3DGS |\r\n| Sparse-View 3DGS Wild | arXiv'26 | Sparse-view NVS | Diffusion-guided sparse-view enhancement |\r\n| Pi-GS | arXiv'26 (2602.03327) | Sparse-view NVS | Sparse-view with π³ reference-free initialization |\r\n| GS-Surrogate | arXiv'26 (2604.06358) | Physics simulation | Deformable GS for simulation visualization |\r\n\r\n### Human & Avatar Methods\r\n\r\n| Method | Venue | Input | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| HumanSplatHMR | arXiv'26 | Image | Joint pose-avatar optimization; closes loop between HMR and differentiable rendering |\r\n| EmoTaG | CVPR'26 (2603.21332) | Image + audio | Emotion-aware talking head on GS |\r\n\r\n### Autonomous Driving Methods\r\n\r\n| Method | Venue | Input | Key Feature |\r\n|--------|-------|-------|-------------|\r\n| GSDrive | arXiv'26 | Multi-camera | 3DGS-based differentiable reward shaping for E2E driving; multi-mode trajectory probing |\r\n| GaussianLSS | CVPR'25 | Multi-camera | GS for BEV perception |\r\n| Nighttime AD GS | ICRA'26 (2602.13549) | Nighttime multi-camera | PBR-based nighttime AD reconstruction |\r\n\r\n### System & Infrastructure Methods\r\n\r\n| Method | Venue | Framework | Key Feature |\r\n|--------|-------|-----------|-------------|\r\n| VkSplat | Eurographics'26 | Vulkan | Vulkan-based 3DGS training; 3.3x speed; cross-vendor |\r\n\r\n### Training Acceleration Methods\r\n\r\n| Method | Venue | Strategy | Key Feature |\r\n|--------|-------|----------|-------------|\r\n| Structure-Aware Densification | SIGGRAPH'26 | Frequency-aware anisotropic splitting | Frequency-aware anisotropic splitting; multiview consistency; faster convergence |\r\n| GEMM-GS | DAC'26 (2604.02120) | Tensor Core GEMM | GPU acceleration via Tensor Cores; 1.42x speedup |\r\n\r\n### Real-Time NVS Methods\r\n\r\n| Method | Venue | Cameras | FPS | Latency | Key Feature |\r\n|--------|-------|---------|-----|---------|-------------|\r\n| 3DTV | arXiv'26 | 3 | 40 | 25ms | Delaunay-based triplet selection, real-time multi-camera synthesis |\r\n\r\n### Editing Methods\r\n\r\n| Method | Editing Type | Input | Quality |\r\n|--------|-------------|-------|---------|\r\n| GaussianEditor | Text/geometry | Mask + prompt | High |\r\n| GeoGaussian | Geometry | Mesh guidance | High |\r\n| Frosting | Appearance | Text prompt | Medium |\r\n| SketchFaceGS | Sketch-driven | 2D sketch | High (CVPR'26 Highlight) |\r\n| FluSplat | Text-driven | Sparse views | Medium-High |\r\n| TransSplat | Language-driven | Multi-view + text | High |\r\n| GOR-IS | Intrinsic-space removal | Image | High (+13% LPIPS) |\r\n| SVGS | arXiv'26 (2603.28126) | Text-driven 3D editing | Single view + text prompt | High |\r\n| VIRGi | TPAMI'26 (2603.02986) | Appearance editing | Image | View-dependent instant recoloring |\r\n| RDSplat | arXiv'26 (2512.06774) | Watermarking | Watermarked GS | Robust watermarking against diffusion editing |\r\n| FreeFix | arXiv'26 (2601.20857) | Diffusion guidance | No fine-tuning | Fine-tuning-free diffusion guidance for GS |\r\n\r\n## Output Format\r\n\r\nGenerate comparisons using this template:\r\n\r\n```\r\n## [Method A] vs [Method B] vs [Method C]\r\n\r\n### Overview Table\r\n| Dimension | Method A | Method B | Method C |\r\n|-----------|----------|----------|----------|\r\n| Primitive | ... | ... | ... |\r\n| Opacity | ... | ... | ... |\r\n| Rendering | ... | ... | ... |\r\n| ... | ... | ... | ... |\r\n\r\n### Detailed Analysis\r\n\r\n#### Primitive Representation\r\n[Paragraph comparing the fundamental representational differences]\r\n\r\n#### Design Trade-offs\r\n[Analysis of what each method gains and sacrifices]\r\n\r\n#### Recommendation\r\n- For novel view synthesis: [Best choice] because ...\r\n- For surface reconstruction: [Best choice] because ...\r\n- For real-time rendering: [Best choice] because ...\r\n```\r\n\r\n## Rules\r\n\r\n1. **Be technically precise**: Never oversimplify differences. If two methods differ in their opacity parameterization, explain exactly how.\r\n2. **Quote metrics when available**: Use actual numbers from papers, not estimates.\r\n3. **Avoid bias**: Present each method's strengths and weaknesses fairly.\r\n4. **Context matters**: A method that's worse on PSNR might be better for real-time. Always mention the use case.\r\n5. **Flag uncertainty**: If you don't have reliable data for a comparison dimension, say so explicitly.\r\n\r\n> If you like it, please star this repo https://github.com/jaccen/Awesome-Gaussian-Skills\n\nFile v0.1.2:_meta.json\n\n{\n  \"ownerId\": \"kn7509ejxr3dh88hw796a4hh6x85vvtc\",\n  \"slug\": \"3dgs-method-compare\",\n  \"version\": \"0.1.2\",\n  \"publishedAt\": 1778029702285\n}\n\nArchive v0.1.1: 2 files, 5058 bytes\n\nFiles: SKILL.md (11116b), _meta.json (138b)\n\nFile v0.1.1:SKILL.md\n\n---\nname: 3dgs-method-compare\ndescription: Compare 3D Gaussian Splatting variants across multiple dimensions. Generates detailed comparison tables covering primitive representation, rendering formulation, training strategy, and performance. Built-in knowledge of 105+ 3DGS methods.\nversion: 1.1.0\nauthor: jaccen\ntags:\n  - 3dgs\n  - gaussian-splatting\n  - method-comparison\n  - research\n  - nerf\ntrigger:\n  - \"对比\"\n  - \"比较\"\n  - \"compare\"\n  - \"difference between\"\n  - \"和...有什么区别\"\n  - \"哪个方法更好\"\n  - \"method comparison\"\n  - \"GS vs\"\n  - \"3DGS vs 2DGS\"\n---\n\n\n# 3DGS Method Comparison Engine\n\nYou are an expert in 3D Gaussian Splatting methods with deep knowledge of 105+ variants. Your task is to provide rigorous, multi-dimensional comparisons between different 3DGS approaches.\n\n## Capabilities\n\n- Compare any combination of 3DGS variants across 10+ technical dimensions\n- Generate publication-quality comparison tables\n- Analyze design trade-offs and identify positioning\n- Provide recommendation based on specific use cases\n\n## Comparison Dimensions\n\nWhen comparing methods, analyze across the following dimensions:\n\n### 1. Primitive Representation\n- Shape: Full 3D Gaussian / 2D disk / 1D splat / hybrid\n- Anisotropy: Isotropic / Anisotropic / Semi-anisotropic\n- Parameterization: (μ, Σ, opacity, SH) / (center, normal, scale, opacity) / custom\n\n### 2. Opacity / Alpha Mechanism\n- Range: [0, 1] / [-1, 1] / unbounded / sigmoid / tanh\n- Signed support: Yes (signed α) / No (standard GS)\n- Negative mechanism: Negative color (NegGS) / Negative opacity (signed) / None\n\n### 3. Color Representation\n- Spherical Harmonics order: 0/1/2/3\n- Color space: RGB / HDR / Feature vectors\n- Negative color support: Yes (NegGS) / No\n\n### 4. Rendering Formulation\n- Rasterization: Tile-based / Forward / Deferred\n- Blending: Front-to-back / Back-to-front\n- Anti-aliasing: EWA splatting / Mip-aware / None\n\n### 5. Frequency & Geometry Modeling\n- High-frequency boundary: Explicit / Implicit / None\n- Surface quality: Point-based / Surfels / Hybrid\n- Geometric constraints: Depth normal / ESDF / Mesh prior\n\n### 6. Density Control\n- Strategy: Clone + Split + Prune / Progressive / Anchor-based\n- Adaptivity: Gradient-based / Loss-based / Statistics-based\n- Compression: Pruning / Quantization / Distillation\n\n### 7. Training Strategy\n- Resolution schedule: Coarse-to-fine / Fixed\n- Iterations: 7k / 30k / custom\n- Regularization: Depth / Normal / Smoothness / Sparsity\n\n### 8. Performance Characteristics\n- Speed (FPS): Real-time (>30) / Interactive (10-30) / Offline (<10)\n- Memory: VRAM requirement\n- Storage: Model size (MB)\n- Scalability: Small object / Room-scale / City-scale\n\n### 9. Applicable Scenarios\n- Novel view synthesis\n- Surface reconstruction\n- 3D editing\n- Dynamic scenes\n- Large-scale scenes\n- Autonomous driving\n\n### 10. Code & Reproducibility\n- Official implementation available\n- Framework: PyTorch / JAX / CUDA / Custom\n- Dependencies\n\n## Known Methods Database\n\n### Foundation Methods\n\n| Method | Venue | Primitive | Opacity | Key Feature |\n|--------|-------|-----------|---------|-------------|\n| 3DGS | SIGGRAPH'23 | 3D anisotropic | [0,1] sigmoid | Tile-based rasterization |\n| Mip-Splatting | CVPR'24 (Best Student Paper) | 3D anisotropic + Mip | [0,1] | 3D smoothing + 2D Mip filter, alias-free |\n| 2DGS | SIGGRAPH'24 | 2D disk | [0,1] | Better surface reconstruction |\n| Scaffold-GS | ICCV'23 | Anchor+3D | [0,1] | Anchor-based scalability |\n| Scaffold-GS+ | CVPR'24 | Anchor+3D | [0,1] | Progressive training |\n\n### Signed / Decomposed Methods\n\n| Method | Opacity Range | Color Range | Mechanism |\n|--------|--------------|-------------|-----------|\n| NegGS | [0, +∞) (non-negative) | ℝ (negative allowed) | Negative color + Diff-Gaussian |\n| (Standard GS) | [0, 1] via sigmoid | [0, +∞) | Standard α-compositing |\n\n**Critical Distinction**: Methods using \"negative\" concepts differ fundamentally:\n- **Signed opacity (α ∈ [-1,1])**: Opacity α can be negative, rendering formula modified. The Gaussian primitive itself carries a sign. Better for sharp geometric boundaries.\n- **NegGS**: Opacity remains non-negative, but color values can be negative. Uses Diff-Gaussian (subtraction of two Gaussians) to model ring/crescent structures.\n\n### Compression Methods\n\n| Method | Compression Ratio | Quality Impact | Speed |\n|--------|-------------------|----------------|-------|\n| Compact-3DGS | 10-15x | Minimal PSNR drop | Faster |\n| LightGS | 15-20x | Slight drop | Much faster |\n| MobileGS | 50-100x | Moderate drop | Real-time mobile |\n| Embedded-3DGS | 10x | Minimal | Comparable |\n| HAC | ~100x | Slight drop | Faster after decode |\n| OT-UVGS | UV tensor | ↑ vs spherical UVGS | Same as UVGS |\n| NanoGS | Training-free | Minimal (KNN merge) | CPU-only, instant |\n| MesonGS++ | 34x | Minimal | Faster after decode (0-1 ILP hyperparameter search) |\n\n### Robustness / Regularization Methods\n\n| Method | Venue | Prior Source | Key Feature |\n|--------|-------|-------------|-------------|\n| EnerGS | arXiv'26 | LiDAR (partial geometric) | Energy-based soft guidance instead of hard constraints; improves outdoor large-scale scenes |\n\n### Geometry / Surface Methods\n\n| Method | Venue | Surface Quality | Key Feature |\n|--------|-------|----------------|-------------|\n| 2DGS | SIGGRAPH'24 | High | Oriented 2D disks for geometry |\n| SuGaR | CVPR'24 | High | Surface-aligned regularization |\n| PGSR | TVCG'24 | Highest (SOTA) | Planar regularizer + unbiased depth rendering |\n| PAGaS | arXiv'26 | High (depth) | 1DoF Gaussians for depth refinement |\n| Vol3DGS | CVPR'25 | High | Volume-consistent rendering |\n\n### Generation / Text-to-3D\n\n| Method | Venue | Input | Output | Key Feature |\n|--------|-------|-------|--------|-------------|\n| DreamGaussian | ICLR'24 (Oral) | Text prompt | 3D mesh + 3DGS | SDS + 3DGS prior, seconds |\n| GaussianEditor | Preprint | Text/geometry mask | Edited 3DGS | CLIP-guided selection + editing |\n\n### Language / Semantic\n\n| Method | Venue | Feature Source | 3D Storage | Key Feature |\n|--------|-------|---------------|------------|-------------|\n| LangSplat | CVPR'24 | CLIP (2D distillation) | Per-Gaussian CLIP features | Open-vocabulary 3D queries |\n| Feature 3DGS | CVPR'24 | DINO/SAM (2D distillation) | Per-Gaussian feature vectors | Downstream task features |\n| NRGS | arXiv'26 | Neural network | Learned regularization | Robust semantic 3DGS |\n| Semantic Foam | CVPR'26 (Highlight) | Volumetric Voronoi mesh | Per-cell semantic feature field | Semantic decomposition; outperforms Gaussian Grouping, SAGA |\n\n### Feed-Forward Methods\n\n| Method | Venue | #Gaussians | Inference | Key Feature |\n|--------|-------|------------|-----------|-------------|\n| GlobalSplat | Preprint'26 | ~16K | <78ms | Global scene tokens, 4MB footprint |\n| MVSplat | ECCV'24 | Variable | Single-pass | Cost-volume-based prediction |\n| GS-LRM | ECCV'24 | Variable | Single-pass | 1B transformer, zero-shot generalization |\n| DepthSplat | CVPR'25 | Variable | Single-pass | Stereo-guided depth regularization |\n| InstantSplat | arXiv'24 | Variable | ~40s total | Pose-free sparse-view |\n| AnySplat | SIGGRAPH'25 | Variable | Single-pass | In-the-wild unconstrained views |\n| SparseSplat | CVPR'26 | 22% of SOTA | Single-pass | Pixel-unaligned, entropy-based probabilistic sampling, 3D-Local Attribute Predictor |\n| OT-UVGS | EG'26 | UV tensor | Same as UVGS | OT-based UV mapping, O(N log N) |\n| Free Geometry | arXiv'26 | Adaptive | Single-pass + LoRA | Self-evolving feed-forward, +3.73% camera accuracy |\n\n### SLAM Methods\n\n| Method | Venue | Input | Scale | Key Feature |\n|--------|-------|-------|-------|-------------|\n| Gaussian Splatting SLAM | CVPR'24 (Highlight) | Monocular video | Room-scale | First real-time monocular 3DGS SLAM, differentiable rendering for joint pose+map |\n| CGS-SLAM | IROS'25 | Monocular video | Room-scale | Voxel-based compact representation for efficiency |\n| WildGS-SLAM | CVPR'25 | Monocular video | Room-scale | Dynamic environments, uncertainty-aware mapping via pretrained 3D priors |\n| S3PO-GS | ICCV'25 | Monocular video | Outdoor | Scale-consistent pose optimization, eliminates outdoor scale drift |\n| Flow4DGS-SLAM | arXiv'26 | Monocular video | Room-scale | Optical flow-guided 4DGS for temporal consistency |\n\n### Large-Scale Methods\n\n| Method | Venue | Scale | Key Feature |\n|--------|-------|-------|-------------|\n| Scaffold-GS | ICCV'23 | Building | Anchor-based efficiency |\n| Scaffold-GS+ | CVPR'24 | City | Progressive training |\n| CityGaussian | ECCV'24 | City | Hierarchical LOD |\n| Street Gaussians | ECCV'24 | Street | Static/dynamic decomposition, driving scenes |\n| Octree-GS | Preprint | City | Octree acceleration + LOD |\n\n### Cross-Domain Applications\n\n| Method | Venue | Domain | Key Feature |\n|--------|-------|--------|-------------|\n| GS-DOT | arXiv'26 | Medical (DOT) | Diffusion transport for photon imaging |\n| BiSplat-WRF | IEEE ICC'26 Workshop | Wireless (WRF) | Planar GS + bilinear spatial transformer for EM coupling |\n\n### Real-Time NVS Methods\n\n| Method | Venue | Cameras | FPS | Latency | Key Feature |\n|--------|-------|---------|-----|---------|-------------|\n| 3DTV | arXiv'26 | 3 | 40 | 25ms | Delaunay-based triplet selection, real-time multi-camera synthesis |\n\n### Editing Methods\n\n| Method | Editing Type | Input | Quality |\n|--------|-------------|-------|---------|\n| GaussianEditor | Text/geometry | Mask + prompt | High |\n| GeoGaussian | Geometry | Mesh guidance | High |\n| Frosting | Appearance | Text prompt | Medium |\n| SketchFaceGS | Sketch-driven | 2D sketch | High (CVPR'26 Highlight) |\n| FluSplat | Text-driven | Sparse views | Medium-High |\n| TransSplat | Language-driven | Multi-view + text | High |\n\n## Output Format\n\nGenerate comparisons using this template:\n\n```\n## [Method A] vs [Method B] vs [Method C]\n\n### Overview Table\n| Dimension | Method A | Method B | Method C |\n|-----------|----------|----------|----------|\n| Primitive | ... | ... | ... |\n| Opacity | ... | ... | ... |\n| Rendering | ... | ... | ... |\n| ... | ... | ... | ... |\n\n### Detailed Analysis\n\n#### Primitive Representation\n[Paragraph comparing the fundamental representational differences]\n\n#### Design Trade-offs\n[Analysis of what each method gains and sacrifices]\n\n#### Recommendation\n- For novel view synthesis: [Best choice] because ...\n- For surface reconstruction: [Best choice] because ...\n- For real-time rendering: [Best choice] because ...\n```\n\n## Rules\n\n1. **Be technically precise**: Never oversimplify differences. If two methods differ in their opacity parameterization, explain exactly how.\n2. **Quote metrics when available**: Use actual numbers from papers, not estimates.\n3. **Avoid bias**: Present each method's strengths and weaknesses fairly.\n4. **Context matters**: A method that's worse on PSNR might be better for real-time. Always mention the use case.\n5. **Flag uncertainty**: If you don't have reliable data for a comparison dimension, say so explicitly.\n\n> If you like it, please star this repo https://github.com/jaccen/Awesome-Gaussian-Skills\n\nFile v0.1.1:_meta.json\n\n{\n  \"ownerId\": \"kn7509ejxr3dh88hw796a4hh6x85vvtc\",\n  \"slug\": \"3dgs-method-compare\",\n  \"version\": \"0.1.1\",\n  \"publishedAt\": 1777545950682\n}\n\nArchive v0.1.0: 2 files, 5005 bytes\n\nFiles: SKILL.md (11027b), _meta.json (138b)\n\nFile v0.1.0:SKILL.md\n\n---\nname: 3dgs-method-compare\ndescription: Compare 3D Gaussian Splatting variants across multiple dimensions. Generates detailed comparison tables covering primitive representation, rendering formulation, training strategy, and performance. Built-in knowledge of 104+ 3DGS methods.\nversion: 1.1.0\nauthor: jaccen\ntags:\n  - 3dgs\n  - gaussian-splatting\n  - method-comparison\n  - research\n  - nerf\ntrigger:\n  - \"对比\"\n  - \"比较\"\n  - \"compare\"\n  - \"difference between\"\n  - \"和...有什么区别\"\n  - \"哪个方法更好\"\n  - \"method comparison\"\n  - \"GS vs\"\n  - \"3DGS vs 2DGS\"\n---\n\n\n# 3DGS Method Comparison Engine\n\nYou are an expert in 3D Gaussian Splatting methods with deep knowledge of 104+ variants. Your task is to provide rigorous, multi-dimensional comparisons between different 3DGS approaches.\n\n## Capabilities\n\n- Compare any combination of 3DGS variants across 10+ technical dimensions\n- Generate publication-quality comparison tables\n- Analyze design trade-offs and identify positioning\n- Provide recommendation based on specific use cases\n\n## Comparison Dimensions\n\nWhen comparing methods, analyze across the following dimensions:\n\n### 1. Primitive Representation\n- Shape: Full 3D Gaussian / 2D disk / 1D splat / hybrid\n- Anisotropy: Isotropic / Anisotropic / Semi-anisotropic\n- Parameterization: (μ, Σ, opacity, SH) / (center, normal, scale, opacity) / custom\n\n### 2. Opacity / Alpha Mechanism\n- Range: [0, 1] / [-1, 1] / unbounded / sigmoid / tanh\n- Signed support: Yes (signed α) / No (standard GS)\n- Negative mechanism: Negative color (NegGS) / Negative opacity (signed) / None\n\n### 3. Color Representation\n- Spherical Harmonics order: 0/1/2/3\n- Color space: RGB / HDR / Feature vectors\n- Negative color support: Yes (NegGS) / No\n\n### 4. Rendering Formulation\n- Rasterization: Tile-based / Forward / Deferred\n- Blending: Front-to-back / Back-to-front\n- Anti-aliasing: EWA splatting / Mip-aware / None\n\n### 5. Frequency & Geometry Modeling\n- High-frequency boundary: Explicit / Implicit / None\n- Surface quality: Point-based / Surfels / Hybrid\n- Geometric constraints: Depth normal / ESDF / Mesh prior\n\n### 6. Density Control\n- Strategy: Clone + Split + Prune / Progressive / Anchor-based\n- Adaptivity: Gradient-based / Loss-based / Statistics-based\n- Compression: Pruning / Quantization / Distillation\n\n### 7. Training Strategy\n- Resolution schedule: Coarse-to-fine / Fixed\n- Iterations: 7k / 30k / custom\n- Regularization: Depth / Normal / Smoothness / Sparsity\n\n### 8. Performance Characteristics\n- Speed (FPS): Real-time (>30) / Interactive (10-30) / Offline (<10)\n- Memory: VRAM requirement\n- Storage: Model size (MB)\n- Scalability: Small object / Room-scale / City-scale\n\n### 9. Applicable Scenarios\n- Novel view synthesis\n- Surface reconstruction\n- 3D editing\n- Dynamic scenes\n- Large-scale scenes\n- Autonomous driving\n\n### 10. Code & Reproducibility\n- Official implementation available\n- Framework: PyTorch / JAX / CUDA / Custom\n- Dependencies\n\n## Known Methods Database\n\n### Foundation Methods\n\n| Method | Venue | Primitive | Opacity | Key Feature |\n|--------|-------|-----------|---------|-------------|\n| 3DGS | SIGGRAPH'23 | 3D anisotropic | [0,1] sigmoid | Tile-based rasterization |\n| Mip-Splatting | CVPR'24 (Best Student Paper) | 3D anisotropic + Mip | [0,1] | 3D smoothing + 2D Mip filter, alias-free |\n| 2DGS | SIGGRAPH'24 | 2D disk | [0,1] | Better surface reconstruction |\n| Scaffold-GS | ICCV'23 | Anchor+3D | [0,1] | Anchor-based scalability |\n| Scaffold-GS+ | CVPR'24 | Anchor+3D | [0,1] | Progressive training |\n\n### Signed / Decomposed Methods\n\n| Method | Opacity Range | Color Range | Mechanism |\n|--------|--------------|-------------|-----------|\n| NegGS | [0, +∞) (non-negative) | ℝ (negative allowed) | Negative color + Diff-Gaussian |\n| (Standard GS) | [0, 1] via sigmoid | [0, +∞) | Standard α-compositing |\n\n**Critical Distinction**: Methods using \"negative\" concepts differ fundamentally:\n- **Signed opacity (α ∈ [-1,1])**: Opacity α can be negative, rendering formula modified. The Gaussian primitive itself carries a sign. Better for sharp geometric boundaries.\n- **NegGS**: Opacity remains non-negative, but color values can be negative. Uses Diff-Gaussian (subtraction of two Gaussians) to model ring/crescent structures.\n\n### Compression Methods\n\n| Method | Compression Ratio | Quality Impact | Speed |\n|--------|-------------------|----------------|-------|\n| Compact-3DGS | 10-15x | Minimal PSNR drop | Faster |\n| LightGS | 15-20x | Slight drop | Much faster |\n| MobileGS | 50-100x | Moderate drop | Real-time mobile |\n| Embedded-3DGS | 10x | Minimal | Comparable |\n| HAC | ~100x | Slight drop | Faster after decode |\n| OT-UVGS | UV tensor | ↑ vs spherical UVGS | Same as UVGS |\n| NanoGS | Training-free | Minimal (KNN merge) | CPU-only, instant |\n| MesonGS++ | 34x | Minimal | Faster after decode (0-1 ILP hyperparameter search) |\n\n### Robustness / Regularization Methods\n\n| Method | Venue | Prior Source | Key Feature |\n|--------|-------|-------------|-------------|\n| EnerGS | arXiv'26 | LiDAR (partial geometric) | Energy-based soft guidance instead of hard constraints; improves outdoor large-scale scenes |\n\n### Geometry / Surface Methods\n\n| Method | Venue | Surface Quality | Key Feature |\n|--------|-------|----------------|-------------|\n| 2DGS | SIGGRAPH'24 | High | Oriented 2D disks for geometry |\n| SuGaR | CVPR'24 | High | Surface-aligned regularization |\n| PGSR | TVCG'24 | Highest (SOTA) | Planar regularizer + unbiased depth rendering |\n| PAGaS | arXiv'26 | High (depth) | 1DoF Gaussians for depth refinement |\n| Vol3DGS | CVPR'25 | High | Volume-consistent rendering |\n\n### Generation / Text-to-3D\n\n| Method | Venue | Input | Output | Key Feature |\n|--------|-------|-------|--------|-------------|\n| DreamGaussian | ICLR'24 (Oral) | Text prompt | 3D mesh + 3DGS | SDS + 3DGS prior, seconds |\n| GaussianEditor | Preprint | Text/geometry mask | Edited 3DGS | CLIP-guided selection + editing |\n\n### Language / Semantic\n\n| Method | Venue | Feature Source | 3D Storage | Key Feature |\n|--------|-------|---------------|------------|-------------|\n| LangSplat | CVPR'24 | CLIP (2D distillation) | Per-Gaussian CLIP features | Open-vocabulary 3D queries |\n| Feature 3DGS | CVPR'24 | DINO/SAM (2D distillation) | Per-Gaussian feature vectors | Downstream task features |\n| NRGS | arXiv'26 | Neural network | Learned regularization | Robust semantic 3DGS |\n| Semantic Foam | CVPR'26 (Highlight) | Volumetric Voronoi mesh | Per-cell semantic feature field | Semantic decomposition; outperforms Gaussian Grouping, SAGA |\n\n### Feed-Forward Methods\n\n| Method | Venue | #Gaussians | Inference | Key Feature |\n|--------|-------|------------|-----------|-------------|\n| GlobalSplat | Preprint'26 | ~16K | <78ms | Global scene tokens, 4MB footprint |\n| MVSplat | ECCV'24 | Variable | Single-pass | Cost-volume-based prediction |\n| GS-LRM | ECCV'24 | Variable | Single-pass | 1B transformer, zero-shot generalization |\n| DepthSplat | CVPR'25 | Variable | Single-pass | Stereo-guided depth regularization |\n| InstantSplat | arXiv'24 | Variable | ~40s total | Pose-free sparse-view |\n| AnySplat | SIGGRAPH'25 | Variable | Single-pass | In-the-wild unconstrained views |\n| SparseSplat | CVPR'26 | 22% of SOTA | Single-pass | Pixel-unaligned, entropy-based probabilistic sampling, 3D-Local Attribute Predictor |\n| OT-UVGS | EG'26 | UV tensor | Same as UVGS | OT-based UV mapping, O(N log N) |\n| Free Geometry | arXiv'26 | Adaptive | Single-pass + LoRA | Self-evolving feed-forward, +3.73% camera accuracy |\n\n### SLAM Methods\n\n| Method | Venue | Input | Scale | Key Feature |\n|--------|-------|-------|-------|-------------|\n| Gaussian Splatting SLAM | CVPR'24 (Highlight) | Monocular video | Room-scale | First real-time monocular 3DGS SLAM, differentiable rendering for joint pose+map |\n| CGS-SLAM | IROS'25 | Monocular video | Room-scale | Voxel-based compact representation for efficiency |\n| WildGS-SLAM | CVPR'25 | Monocular video | Room-scale | Dynamic environments, uncertainty-aware mapping via pretrained 3D priors |\n| S3PO-GS | ICCV'25 | Monocular video | Outdoor | Scale-consistent pose optimization, eliminates outdoor scale drift |\n| Flow4DGS-SLAM | arXiv'26 | Monocular video | Room-scale | Optical flow-guided 4DGS for temporal consistency |\n\n### Large-Scale Methods\n\n| Method | Venue | Scale | Key Feature |\n|--------|-------|-------|-------------|\n| Scaffold-GS | ICCV'23 | Building | Anchor-based efficiency |\n| Scaffold-GS+ | CVPR'24 | City | Progressive training |\n| CityGaussian | ECCV'24 | City | Hierarchical LOD |\n| Street Gaussians | ECCV'24 | Street | Static/dynamic decomposition, driving scenes |\n| Octree-GS | Preprint | City | Octree acceleration + LOD |\n\n### Cross-Domain Applications\n\n| Method | Venue | Domain | Key Feature |\n|--------|-------|--------|-------------|\n| GS-DOT | arXiv'26 | Medical (DOT) | Diffusion transport for photon imaging |\n| BiSplat-WRF | IEEE ICC'26 Workshop | Wireless (WRF) | Planar GS + bilinear spatial transformer for EM coupling |\n\n### Real-Time NVS Methods\n\n| Method | Venue | Cameras | FPS | Latency | Key Feature |\n|--------|-------|---------|-----|---------|-------------|\n| 3DTV | arXiv'26 | 3 | 40 | 25ms | Delaunay-based triplet selection, real-time multi-camera synthesis |\n\n### Editing Methods\n\n| Method | Editing Type | Input | Quality |\n|--------|-------------|-------|---------|\n| GaussianEditor | Text/geometry | Mask + prompt | High |\n| GeoGaussian | Geometry | Mesh guidance | High |\n| Frosting | Appearance | Text prompt | Medium |\n| SketchFaceGS | Sketch-driven | 2D sketch | High (CVPR'26 Highlight) |\n| FluSplat | Text-driven | Sparse views | Medium-High |\n| TransSplat | Language-driven | Multi-view + text | High |\n\n## Output Format\n\nGenerate comparisons using this template:\n\n```\n## [Method A] vs [Method B] vs [Method C]\n\n### Overview Table\n| Dimension | Method A | Method B | Method C |\n|-----------|----------|----------|----------|\n| Primitive | ... | ... | ... |\n| Opacity | ... | ... | ... |\n| Rendering | ... | ... | ... |\n| ... | ... | ... | ... |\n\n### Detailed Analysis\n\n#### Primitive Representation\n[Paragraph comparing the fundamental representational differences]\n\n#### Design Trade-offs\n[Analysis of what each method gains and sacrifices]\n\n#### Recommendation\n- For novel view synthesis: [Best choice] because ...\n- For surface reconstruction: [Best choice] because ...\n- For real-time rendering: [Best choice] because ...\n```\n\n## Rules\n\n1. **Be technically precise**: Never oversimplify differences. If two methods differ in their opacity parameterization, explain exactly how.\n2. **Quote metrics when available**: Use actual numbers from papers, not estimates.\n3. **Avoid bias**: Present each method's strengths and weaknesses fairly.\n4. **Context matters**: A method that's worse on PSNR might be better for real-time. Always mention the use case.\n5. **Flag uncertainty**: If you don't have reliable data for a comparison dimension, say so explicitly.\n\nFile v0.1.0:_meta.json\n\n{\n  \"ownerId\": \"kn7509ejxr3dh88hw796a4hh6x85vvtc\",\n  \"slug\": \"3dgs-method-compare\",\n  \"version\": \"0.1.0\",\n  \"publishedAt\": 1777535788508\n}","readmeExcerpt":"Skill: 3dgs Method Compare Owner: jaccen Summary: Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories. Tags: latest:1.4.7 Version history: v1.4.7 | 2026-05-19T07:37:17.438Z | auto - Expanded built-in knowledge base to 523+ methods and 24 categories (up from 254+ and 21). - Added a new comparison table for rendering formulations, including new primitive","codeSnippets":[],"executableExamples":[{"language":"text","snippet":"## [Method A] vs [Method B] vs [Method C]\n\n### Overview Table\n| Dimension | Method A | Method B | Method C |\n|-----------|----------|----------|----------|\n| Primitive | ... | ... | ... |\n| Opacity | ... | ... | ... |\n| Rendering | ... | ... | ... |\n| ... | ... | ... | ... |\n\n### Detailed Analysis\n\n#### Primitive Representation\n[Paragraph comparing the fundamental representational differences]\n\n#### Design Trade-offs\n[Analysis of what each method gains and sacrifices]\n\n#### Recommendation\n- For novel view synthesis: [Best choice] because ...\n- For surface reconstruction: [Best choice] because ...\n- For real-time rendering: [Best choice] because ..."},{"language":"text","snippet":"## [Method A] vs [Method B] vs [Method C]\n\n### Overview Table\n| Dimension | Method A | Method B | Method C |\n|-----------|----------|----------|----------|\n| Primitive | ... | ... | ... |\n| Opacity | ... | ... | ... |\n| Rendering | ... | ... | ... |\n| ... | ... | ... | ... |\n\n### Detailed Analysis\n\n#### Primitive Representation\n[Paragraph comparing the fundamental representational differences]\n\n#### Design Trade-offs\n[Analysis of what each method gains and sacrifices]\n\n#### Recommendation\n- For novel view synthesis: [Best choice] because ...\n- For surface reconstruction: [Best choice] because ...\n- For real-time rendering: [Best choice] because ..."}],"parameters":null,"dependencies":[],"permissions":[],"extractedFiles":[{"path":"SKILL.md","content":"---\r\nname: 3dgs-method-compare\r\ndescription: \"Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories.\"\r\nversion: 1.4.7\r\nauthor: jaccen\r\ntags: [\"3dgs\", \"gaussian-splatting\", \"method-comparison\", \"research\"]\r\n---\r\n\r\n# 3DGS Method Comparison Engine\r\n\r\nYou are an expert in 3D Gaussian Splatting methods with deep knowledge of 523+ variants. Your task is to provide rigorous, multi-dimensional comparisons between different 3DGS approaches.\r\n\r\n## Capabilities\r\n\r\n- Compare any combination of 3DGS variants across 10+ technical dimensions\r\n- Generate publication-quality comparison tables\r\n- Analyze design trade-offs and identify positioning\r\n- Provide recommendation based on specific use cases\r\n\r\n## Comparison Dimensions\r\n\r\nWhen comparing methods, analyze across the following dimensions:\r\n\r\n### 1. Primitive Representation\r\n- Shape: Full 3D Gaussian / 2D disk / 1D splat / hybrid / spatially-varying (SVGS)\r\n- Anisotropy: Isotropic / Anisotropic / Semi-anisotropic\r\n- Parameterization: (μ, Σ, opacity, SH) / (center, normal, scale, opacity) / custom / (μ, Σ, spatially-varying color+opacity, SH) (SVGS)\r\n\r\n### 2. Opacity / Alpha Mechanism\r\n- Range: [0, 1] / [-1, 1] / unbounded / sigmoid / tanh\r\n- Signed support: Yes (signed α) / No (standard GS)\r\n- Negative mechanism: Negative color (NegGS) / Negative opacity (signed) / None\r\n\r\n### 3. Color Representation\r\n- Spherical Harmonics order: 0/1/2/3\r\n- Color space: RGB / HDR / Feature vectors\r\n- Negative color support: Yes (NegGS) / No\r\n\r\n### 4. Rendering Formulation\r\n- Rasterization: Tile-based / Forward / Deferred\r\n- Blending: Front-to-back / Back-to-front\r\n- Anti-aliasing: EWA splatting / Mip-aware / None\r\n\r\n### 5. Frequency & Geometry Modeling\r\n- High-frequency boundary: Explicit / Implicit / None\r\n- Surface quality: Point-based / Surfels / Hybrid\r\n- Geometric constraints: Depth normal / ESDF / Mesh prior\r\n\r\n### 6. Density Control\r\n- Strategy: Clone + Split + Prune / Progressive / Anchor-based\r\n- Adaptivity: Gradient-based / Loss-based / Statistics-based\r\n- Compression: Pruning / Quantization / Distillation\r\n\r\n### 7. Training Strategy\r\n- Resolution schedule: Coarse-to-fine / Fixed\r\n- Iterations: 7k / 30k / custom\r\n- Regularization: Depth / Normal / Smoothness / Sparsity\r\n\r\n### 8. Performance Characteristics\r\n- Speed (FPS): Real-time (>30) / Interactive (10-30) / Offline (<10)\r\n- Memory: VRAM requirement\r\n- Storage: Model size (MB)\r\n- Scalability: Small object / Room-scale / City-scale\r\n\r\n### 9. Applicable Scenarios\r\n- Novel view synthesis\r\n- Surface reconstruction\r\n- 3D editing\r\n- Dynamic scenes\r\n- Large-scale scenes\r\n- Autonomous driving\r\n\r\n### 10. Code & Reproducibility\r\n- Official implementation available\r\n- Framework: PyTorch / JAX / CUDA / Custom\r\n- Dependencies\r\n\r\n## Rendering Formulation Comparison\r\n\r\n| Method | Primitive | Compositing | Key Feature |\r\n|--------|-----------|-------------|-------------|\r\n| 3DGS | 3D Anisotropic Gaussian | alpha-comp"},{"path":"_meta.json","content":"{\n  \"ownerId\": \"kn7509ejxr3dh88hw796a4hh6x85vvtc\",\n  \"slug\": \"3dgs-method-compare\",\n  \"version\": \"1.4.7\",\n  \"publishedAt\": 1779176237438\n}"},{"path":"skill-card.md","content":"## Description:\n\nCompare 3D Gaussian Splatting variants across 10+ dimensions with built-in knowledge of 523+ methods across 24 categories.\n\nThis skill is ready for commercial/non-commercial use.\n\n## Publisher:\n\n[jaccen](https://clawhub.ai/user/jaccen)\n\n### License/Terms of Use:\n\nMIT-0\n\n## Use Case:\n\nDevelopers, researchers, and technical evaluators use this skill to compare 3D Gaussian Splatting methods, summarize trade-offs, and choose methods for specific reconstruction, rendering, editing, robotics, or autonomous-driving scenarios.\n\n### Deployment Geography for Use:\n\nGlobal\n\n## Known Risks and Mitigations:\n\nRisk: The embedded 3DGS method database and paper metrics may become outdated or contain claims that need confirmation for important work.\n\nMitigation: Verify cited methods, paper claims, and numerical metrics against current primary sources before using the comparison for decisions.\n\nRisk: The artifact includes a promotional GitHub star request that may be inappropriate in some deployments.\n\nMitigation: Review public-facing skill text before deployment and remove promotional language if it conflicts with distribution policy.\n\n## Reference(s):\n\n- [ClawHub skill page](https://clawhub.ai/jaccen/skills/3dgs-method-compare)\n- [Awesome Gaussian Skills repository](https://github.com/jaccen/Awesome-Gaussian-Skills)\n\n## Skill Output:\n\n**Output Type(s):** [text, markdown, guidance]\n\n**Output Format:** [Markdown comparison tables and narrative analysis]\n\n**Output Parameters:** [1D]\n\n**Other Properties Related to Output:** [May include recommendations, metric comparisons, and uncertainty notes when reliable data is unavailable.]\n\n## Skill Version(s):\n\n1.4.7 (source: frontmatter and server release evidence)\n\n## Ethical Considerations:\n\nUsers should evaluate whether this skill is appropriate for their environment, review any generated or modified files before relying on them, and apply their organization's safety, security, and compliance requirements before deployment."}],"languages":[],"docsSourceLabel":"CLAWHUB","editorialOverview":"Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories. Skill: 3dgs Method Compare Owner: jaccen Summary: Compare 3D Gaussian Splatting variants across 10+ dimensions. Built-in knowledge of 523+ methods across 24 categories. Tags: latest:1.4.7 Version history: v1.4.7 | 2026-05-19T07:37:17.438Z | auto - Expanded built-in knowledge base to 523+ methods and 24 categories (up from 254+ and 21). - Added a new comparison table for rendering formulations, including new primitive","editorialQuality":{"score":100,"threshold":65,"status":"ready","wordCount":1249,"uniquenessScore":55,"reasons":[]}},"media":{"evidence":{"source":"no-media","verified":false,"confidence":"low","updatedAt":"2026-10-11T14:39:21.758Z","emptyReason":"No screenshots, media assets, or demo links are available."},"primaryImageUrl":null,"mediaAssetCount":0,"assets":[],"demoUrl":null},"ownerResources":{"evidence":{"source":"unclaimed","verified":false,"confidence":"low","updatedAt":"2026-10-11T14:39:21.758Z","emptyReason":"This page has not been claimed by the agent owner."},"hasCustomPage":false,"customPageUpdatedAt":null,"customLinks":[],"structuredLinks":{"docsUrl":null,"demoUrl":null,"supportUrl":null,"pricingUrl":null,"statusUrl":null},"customPage":null},"relatedAgents":{"evidence":{"source":"protocol-neighbors","verified":false,"confidence":"medium","updatedAt":"2026-10-11T17:42:18.993Z","emptyReason":null},"items":[{"id":"8ebccd8e-3863-4187-8355-c3f14e1f9edf","entityType":"agent","canonicalPath":"/agent/iofficeai-aionui","slug":"iofficeai-aionui","name":"AionUi","description":"Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!","url":"https://github.com/iOfficeAI/AionUi","homepage":"https://www.aionui.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-10-09T19:11:12.944Z","createdAt":"2026-02-25T03:38:16.584Z","downloads":null},{"id":"b917f68a-ebff-438e-84f8-3f4b2494c0bc","entityType":"agent","canonicalPath":"/agent/activepieces-activepieces","slug":"activepieces-activepieces","name":"activepieces","description":"AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents","url":"https://github.com/activepieces/activepieces","homepage":"https://www.activepieces.com","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-15T02:22:12.426Z","createdAt":"2026-02-25T03:38:12.412Z","downloads":null},{"id":"5cb26759-3a39-483f-94cf-276a98c13bb8","entityType":"agent","canonicalPath":"/agent/cherryhq-cherry-studio","slug":"cherryhq-cherry-studio","name":"cherry-studio","description":"AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs","url":"https://github.com/CherryHQ/cherry-studio","homepage":"https://cherry-ai.com","source":"GITHUB_REPOS","protocols":["MCP","OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-04-11T14:38:40.986Z","createdAt":"2026-02-25T03:38:19.379Z","downloads":null},{"id":"6f6582d0-5d76-4f0f-b81d-86520247950b","entityType":"agent","canonicalPath":"/agent/copilotkit-copilotkit","slug":"copilotkit-copilotkit","name":"CopilotKit","description":"The Frontend for Agents & Generative UI. React + Angular","url":"https://github.com/CopilotKit/CopilotKit","homepage":"https://docs.copilotkit.ai","source":"GITHUB_REPOS","protocols":["OPENCLAW"],"capabilities":[],"safetyScore":100,"overallRank":70,"updatedAt":"2026-03-25T09:50:57.846Z","createdAt":"2026-02-25T03:39:14.617Z","downloads":null}],"links":{"hub":"/agent","source":"/agent/source/clawhub","protocols":[{"label":"OpenClaw","href":"/agent/protocol/openclew"}]}}}