--- name: 3dgs-compression-deploy description: "3DGS compression-to-deployment pipeline: quantization (scalar/VQ/mixed-precision), pruning (coreset/adaptive/variational/merge/Bayesian), progressive streaming & LoD, Web/WebGPU/mobile deployment, hardware acceleration (Tensor Core/GEMM/FPGA/ASIC), training-free semantic compression. Covers 53+ methods across 6 compression categories. Use when: compressing 3DGS models, deploying 3DGS to web/mobile/edge, selecting quantization bit-width, designing streaming pipelines, 3DGS压缩/量化/剪枝/部署/流式传输/移动端/硬件加速." license: Apache-2.0 user-invocable: true metadata: version: "1.1.1" author: jaccen tags: ["3dgs", "gaussian-splatting", "compression", "quantization", "pruning", "vector-quantization", "streaming", "deployment", "mobile", "webgpu", "tensor-core", "hardware-acceleration", "bayesian", "semantic-compression"] when_to_use: - "Compress a 3DGS model for storage or transmission" - "Deploy 3DGS to web browser, mobile device, or edge hardware" - "Select quantization scheme, bit-width, or pruning strategy" - "Design progressive streaming or Level-of-Detail pipeline" - "Evaluate hardware acceleration options (Tensor Core, GEMM, FPGA, ASIC)" - "3DGS压缩 / 量化 / 剪枝 / 部署 / 流式传输 / 移动端 / 硬件加速" ---

# 3DGS Compression & Deployment

> End-to-end pipeline from raw 3DGS model to deployed application. Covers 6 compression categories + 4 deployment targets + hardware acceleration.

## Capabilities

- Analyze 3DGS model attributes (position, SH, opacity, scale, rotation) and recommend compression strategy - Select quantization method and bit-width per attribute (scalar, VQ, mixed-precision) - Design pruning pipeline (coreset, adaptive, variational, merge-based) - Plan VQ codebook architecture and residual coding - Architect progressive streaming and LoD systems (static and 4D dynamic) - Guide platform-specific deployment (WebGL, WebGPU, iOS/Android, desktop) - Evaluate hardware acceleration paths (Tensor Core, GEMM, FPGA, ASIC) - Estimate compression ratio, quality loss, and rendering speed for each method combination

## Compression Pipeline

``` Raw 3DGS Model │ ▼ [Step 1] Analysis ── attribute profiling, bottleneck identification │ ▼ [Step 2] Strategy Selection ── target platform → compression recipe │ ▼ [Step 3] Pruning ── reduce Gaussian count (coreset / adaptive / variational / merge) │ ▼ [Step 4] Quantization ── reduce per-attribute bit-width (scalar / VQ / mixed-precision) │ ▼ [Step 5] Vector Quantization ── codebook-based attribute compression (optional, replaces/augments Step 4) │ ▼ [Step 6] Streaming & LoD ── progressive loading structure for network delivery │ ▼ [Step 7] Deployment ── platform-specific renderer and runtime │ ▼ Deployed Application (Web / Mobile / Desktop / Edge) ```

## Step 1: Analysis

Profile the 3DGS model before selecting compression methods:

| Attribute | FP32 Size | Typical Range | Sensitivity to Quantization | |-----------|-----------|---------------|---------------------------| | Position (μ) | 12B/Gaussian | Scene bounds | High — direct geometry impact | | SH (degree 0–3) | 48B/Gaussian | [-1, 1] per coeff | Medium-High — visual quality driver | | Opacity (α) | 4B/Gaussian | [0, 1] | Medium — pruning signal | | Scale (s) | 12B/Gaussian | [1e-5, 1e2] | Medium — anisotropy sensitive | | Rotation (q) | 16B/Gaussian | Unit quaternion | Low-Medium — can tolerate 8-bit |

**Profiling checklist:** 1. Total Gaussian count N and file size S 2. Target platform constraints (memory budget, bandwidth, GPU capability) 3. Quality floor (minimum acceptable PSNR/SSIM) 4. Required FPS threshold 5. Whether dynamic (4DGS) or static scene

## Step 2: Compression Strategy Selection

Decision tree by target platform:

``` Target Platform? ├── Web (WebGL/WebGPU) │ ├── Bandwidth-limited → Pruning + VQ + Streaming (CAGS/HGS pipeline) │ └── Compute-limited → Aggressive pruning + low SH degree + Flux-GS ├── Mobile (iOS/Android) │ ├── Real-time required → Mobile-GS pipeline (depth-aware OIT + distillation + pruning) │ └── Quality priority → MesonGS++ (mixed-precision, budget-controlled) + NanoGS merge ├── Desktop (GPU ≥ RTX 3060) │ ├── Max quality → Light quantization only (8-10 bit, ContextGS entropy coding) │ └── Large scene → Pruning + Streaming + HiGS hierarchical tiles └── Edge / Embedded ├── FPGA targeted → SpqGS (hardware-friendly quantization) + Axis-Shared Accelerator └── Low-power GPU → VEDAL pruning + 4-6 bit quantization + PocketGS on-device ```

**Combined target table:**

| Target | Typical Gaussian Budget | Bit-width Range | Streaming | Key Methods | |--------|------------------------|-----------------|-----------|-------------| | Web | 100K–500K | 4–8 bit | Required | CAGS, StreamLoD-GS, Spark 2.0 | | Mobile | 50K–200K | 4–8 bit | Optional | Mobile-GS, Flux-GS, PocketGS | | Desktop | 500K–5M | 8–16 bit | For large scenes | MesonGS++, HiGS, gsplat | | Edge/FPGA | 10K–100K | 2–6 bit | Required | SpqGS, VEDAL, GEMM-GS |

## Step 3: Quantization

### Method Selection

| Method | Type | Venue | Bit-width | Key Feature | |--------|------|-------|-----------|-------------| | MesonGS++ | Mixed-precision (post-training) | arXiv 2026 | 4–16 bit per attribute | 0-1 ILP hyperparameter search, 34x compression | | GETA-3DGS | Joint pruning + quantization | arXiv 2026 | 4–8 bit heterogeneous | Render-aware saliency, QADG dependency graph | | GSQ | Learned step size | CVPR 2025 | 4–8 bit | Group-wise quantization with learnable step | | ContextGS | Context-model entropy coding | NeurIPS 2024 | 8–16 bit | Anchor-level context replaces uniform quant | | SpqGS | Scalable parallel | CVPR 2025 | Hardware-friendly | Parallel bit allocation for FPGA/ASIC | | SOG-GS | Channel-grouped | CVPR 2025 | Per-channel | Preserves inter-Gaussian correlations | | ZipGS | Pruning + quant + entropy | CVPR 2025 | Variable | Volumetric entropy coding | | GaussianCodec | Entropy-constrained | CVPR 2025 | Rate-distortion optimized | Learned codec with ECVQ | | EAGLES | Quantized embeddings | ECCV 2024 | 8–16 bit | Coarse-to-fine training + pruning | | TC-GS | Tri-plane representation | IEEE 2026 | Implicit via tri-plane | Replaces per-Gaussian SH with shared tri-plane |

### Bit-width Selection Guide

| Attribute | 4–5 bit | 6–8 bit | 8–12 bit | 12–16 bit | |-----------|---------|---------|----------|-----------| | Position | Edge only — visible artifacts | Mobile/Web acceptable | Desktop recommended | Lossless-range | | SH (dc) | Not recommended | Edge/mobile | Desktop | High-fidelity | | SH (rest) | Aggressive mobile | Mobile/Web | Desktop | Unnecessary | | Opacity | Acceptable (post-sigmoid) | Recommended | Overkill | Overkill | | Scale | Log-space 4-bit risky | Log-space 6–8 bit | Recommended | Overkill | | Rotation | 8-bit often sufficient | Standard | Unnecessary | Overkill |

**Rule of thumb:** Position and SH dominate quality; allocate more bits there. Opacity and rotation tolerate aggressive quantization.

## Step 4: Pruning

### Strategies

| Strategy | Method | Venue | Compression | Quality Impact | |----------|--------|-------|-------------|----------------| | **Coreset-based** | Provable Pruning via Coresets | arXiv 2026 | Theoretical guarantee | Minimal — multiplicative approximation | | **Bayesian** | DP-Splat | arXiv 2026 | Automatic complexity control | Minimal — DP prior converges to optimal count | | **Training-free semantic** | CoSAG | arXiv 2026 | 37–76× over LangSplatV2 | Minimal — zero fine-tuning, leverages CLIP features | | **Importance-based** | Prune Wisely (DoG) | CVPR 2026 | 90% reduction | Minimal — DoG avoids false positives | | **Variational** | VEDAL | arXiv 2026（venue 待核实） | 5.2x (0.31 dB drop) | Low — uncertainty-gated async pruning | | **Merge-based** | NanoGS | arXiv 2026 | Training-free | Mass-preserving moment matching | | **Global+Local** | LightGaussian | NeurIPS 2024 | 15x | SVD distillation compensates | | **Render-aware** | GETA-3DGS | arXiv 2026 | ~5x storage | Transmittance-weighted saliency | | **Frequency-aware** | FAD-GS | CVPR 2024 | Frequency-separated | Separates low/high freq Gaussians | | **Memory-bounded** | Gaussians on a Diet | arXiv 2026 | 80% peak memory | Iterative growth+pruning | | **Hybrid** | HybridGS | CVPR 2025 | Explicit+implicit | Neural coding recovers pruned info | | **Budget-controlled** | MGS (Matryoshka) | arXiv 2026 | Continuous LoD | Any prefix of ordered set is coherent |

### Pruning Decision Flow

``` Need theoretical guarantees? ├── Yes → Provable Pruning via Coresets └── No ├── Training-free requirement? │ ├── Yes → NanoGS (merge) or LightGaussian (post-training) │ └── No │ ├── Can retrain/fine-tune after pruning? │ │ ├── Yes → Prune Wisely (DoG) + finetune, or VEDAL │ │ └── No → NanoGS or GETA-3DGS (auto, no per-scene thresholds) │ └── Need continuous quality levels? │ └── Yes → MGS (Matryoshka stochastic budget training) ```

## Step 5: Vector Quantization

### VQ Pipeline

``` Gaussian Attributes │ ▼ [1] Attribute Grouping ── group by type (position, SH, scale/rotation) │ ▼ [2] Codebook Learning ── K-means / learned / residual codebook │ ▼ [3] Assignment ── nearest-neighbor lookup per group │ ▼ [4] Residual Coding ── (optional) multi-level residual VQ │ ▼ [5] Entropy Coding ── arithmetic / ANS coding of indices │ ▼ Compressed Bitstream ```

### VQ Methods

| Method | Codebook Type | Venue | Compression | Key Feature | |--------|--------------|-------|-------------|-------------| | CompactGS | Learned per-attribute | ECCV 2024 | 10–15x | Simple codebook, minimal overhead | | VQGS | Residual codebook | CVPR 2025 | High-ratio | Multi-level residual improves quality | | RDO-Gaussian | ECVQ (entropy-constrained) | ECCV 2024 | 40x+ | Rate-distortion optimized VQ | | CAGS | VQ + LoD layers | SIGGRAPH 2026 | Adaptive | VQ establishes quality LoDs for streaming | | CGVQ | Clustered codebook | SIGGRAPH 2026 Poster | 20% bpp reduction | Cluster-guided grouping before quant | | Sp2403GS | Codebook + pruning | CVPR 2024 | Combined | Importance-based codebook selection | | HAC | Hash-grid context | ECCV 2024 | ~100x | Context modeling for entropy coding | | CompGS | Importance-aware | CVPR 2025 | Progressive | Progressive decoding support |

### Codebook Design Rules

1. **Codebook size K**: 256 (8-bit index) is standard; 1024 (10-bit) for quality; 64 (6-bit) for extreme compression 2. **Grouping strategy**: Group by attribute type (position separate from SH); within SH, separate DC from higher-order 3. **Residual levels**: 1 level = 10–20x; 2 levels = 20–50x; 3 levels = diminishing returns 4. **LoD integration**: Each residual level can serve as a LoD tier (see Step 6)

## Step 6: Streaming & LoD

### Static Scene Streaming

| Method | LoD Mechanism | Venue | Latency Reduction | Key Feature | |--------|--------------|-------|-------------------|-------------| | StreamLoD-GS | View-dependent LoD levels | arXiv 2026 | Progressive | Bandwidth-adaptive FVV delivery | | HGS | Hierarchical Gaussian structuring | CVPR 2025 | Progressive | Level-of-detail Gaussian hierarchy | | GS-Stream | Progressive chunk delivery | CVPR 2025 | Bandwidth-adaptive | Chunk-based 3DGS streaming | | EvoGS | Evolution Tree (wavelet-inspired) | arXiv 2026 | 2.4x payload reduction | Continuous parent-child refinement | | MGS | Stochastic budget training | arXiv 2026 | Continuous | Any prefix = coherent render | | SCube | VoxSplats + hierarchical LOD | NeurIPS 2024 | Hierarchical | Voxelized splat for large-scale | | CAGS | VQ-based LoD + reference image | SIGGRAPH 2026 | +5–20 dB PSNR | Server-side low-res reference corrects color |

### Dynamic (4DGS) Streaming

| Method | Mechanism | Venue | First-frame Latency | Key Feature | |--------|-----------|-------|---------------------|-------------|

--- name: 3dgs-compression-deploy description: "3DGS compression-to-deployment pipeline: quantization (scalar/VQ/mixed-precision), pruning (coreset/adaptive/variational/merge/Bayesian), progressive streaming & LoD, Web/WebGPU/mobile deployment, hardware acceleration (Tensor Core/GEMM/FPGA/ASIC), training-free semantic compression. Covers 53+ methods across 6 compression categories. Use when: compressing 3DGS models, deploying 3DGS to web/mobile/edge, selecting quantization bit-width, designing streaming pipelines, 3DGS压缩/量化/剪枝/部署/流式传输/移动端/硬件加速." license: Apache-2.0 user-invocable: true metadata: version: "1.1.1" author: jaccen tags: ["3dgs", "gaussian-splatting", "compression", "quantization", "pruning", "vector-quantization", "streaming", "deployment", "mobile", "webgpu", "tensor-core", "hardware-acceleration", "bayesian", "semantic-compression"] when_to_use: - "Compress a 3DGS model for storage or transmission" - "Deploy 3DGS to web browser, mobile device, or edge hardware" - "Select quantization scheme, bit-width, or pruning strategy" - "Design progressive streaming or Level-of-Detail pipeline" - "Evaluate hardware acceleration options (Tensor Core, GEMM, FPGA, ASIC)" - "3DGS压缩 / 量化 / 剪枝 / 部署 / 流式传输 / 移动端 / 硬件加速" ---

# 3DGS Compression & Deployment

> End-to-end pipeline from raw 3DGS model to deployed application. Covers 6 compression categories + 4 deployment targets + hardware acceleration.

## Capabilities

- Analyze 3DGS model attributes (position, SH, opacity, scale, rotation) and recommend compression strategy - Select quantization method and bit-width per attribute (scalar, VQ, mixed-precision) - Design pruning pipeline (coreset, adaptive, variational, merge-based) - Plan VQ codebook architecture and residual coding - Architect progressive streaming and LoD systems (static and 4D dynamic) - Guide platform-specific deployment (WebGL, WebGPU, iOS/Android, desktop) - Evaluate hardware acceleration paths (Tensor Core, GEMM, FPGA, ASIC) - Estimate compression ratio, quality loss, and rendering speed for each method combination

## Compression Pipeline

``` Raw 3DGS Model │ ▼ [Step 1] Analysis ── attribute profiling, bottleneck identification │ ▼ [Step 2] Strategy Selection ── target platform → compression recipe │ ▼ [Step 3] Pruning ── reduce Gaussian count (coreset / adaptive / variational / merge) │ ▼ [Step 4] Quantization ── reduce per-attribute bit-width (scalar / VQ / mixed-precision) │ ▼ [Step 5] Vector Quantization ── codebook-based attribute compression (optional, replaces/augments Step 4) │ ▼ [Step 6] Streaming & LoD ── progressive loading structure for network delivery │ ▼ [Step 7] Deployment ── platform-specific renderer and runtime │ ▼ Deployed Application (Web / Mobile / Desktop / Edge) ```

## Step 1: Analysis

Profile the 3DGS model before selecting compression methods:

| Attribute | FP32 Size | Typical Range | Sensitivity to Quantization | |-----------|-----------|---------------|---------------------------| | Position (μ) | 12B/Gaussian | Scene bounds | High — direct geometry impact | | SH (degree 0–3) | 48B/Gaussian | [-1, 1] per coeff | Medium-High — visual quality driver | | Opacity (α) | 4B/Gaussian | [0, 1] | Medium — pruning signal | | Scale (s) | 12B/Gaussian | [1e-5, 1e2] | Medium — anisotropy sensitive | | Rotation (q) | 16B/Gaussian | Unit quaternion | Low-Medium — can tolerate 8-bit |

**Profiling checklist:** 1. Total Gaussian count N and file size S 2. Target platform constraints (memory budget, bandwidth, GPU capability) 3. Quality floor (minimum acceptable PSNR/SSIM) 4. Required FPS threshold 5. Whether dynamic (4DGS) or static scene

## Step 2: Compression Strategy Selection

Decision tree by target platform:

``` Target Platform? ├── Web (WebGL/WebGPU) │ ├── Bandwidth-limited → Pruning + VQ + Streaming (CAGS/HGS pipeline) │ └── Compute-limited → Aggressive pruning + low SH degree + Flux-GS ├── Mobile (iOS/Android) │ ├── Real-time required → Mobile-GS pipeline (depth-aware OIT + distillation + pruning) │ └── Quality priority → MesonGS++ (mixed-precision, budget-controlled) + NanoGS merge ├── Desktop (GPU ≥ RTX 3060) │ ├── Max quality → Light quantization only (8-10 bit, ContextGS entropy coding) │ └── Large scene → Pruning + Streaming + HiGS hierarchical tiles └── Edge / Embedded ├── FPGA targeted → SpqGS (hardware-friendly quantization) + Axis-Shared Accelerator └── Low-power GPU → VEDAL pruning + 4-6 bit quantization + PocketGS on-device ```

**Combined target table:**

| Target | Typical Gaussian Budget | Bit-width Range | Streaming | Key Methods | |--------|------------------------|-----------------|-----------|-------------| | Web | 100K–500K | 4–8 bit | Required | CAGS, StreamLoD-GS, Spark 2.0 | | Mobile | 50K–200K | 4–8 bit | Optional | Mobile-GS, Flux-GS, PocketGS | | Desktop | 500K–5M | 8–16 bit | For large scenes | MesonGS++, HiGS, gsplat | | Edge/FPGA | 10K–100K | 2–6 bit | Required | SpqGS, VEDAL, GEMM-GS |

## Step 3: Quantization

### Method Selection

| Method | Type | Venue | Bit-width | Key Feature | |--------|------|-------|-----------|-------------| | MesonGS++ | Mixed-precision (post-training) | arXiv 2026 | 4–16 bit per attribute | 0-1 ILP hyperparameter search, 34x compression | | GETA-3DGS | Joint pruning + quantization | arXiv 2026 | 4–8 bit heterogeneous | Render-aware saliency, QADG dependency graph | | GSQ | Learned step size | CVPR 2025 | 4–8 bit | Group-wise quantization with learnable step | | ContextGS | Context-model entropy coding | NeurIPS 2024 | 8–16 bit | Anchor-level context replaces uniform quant | | SpqGS | Scalable parallel | CVPR 2025 | Hardware-friendly | Parallel bit allocation for FPGA/ASIC | | SOG-GS | Channel-grouped | CVPR 2025 | Per-channel | Preserves inter-Gaussian correlations | | ZipGS | Pruning + quant + entropy | CVPR 2025 | Variable | Volumetric entropy coding | | GaussianCodec | Entropy-constrained | CVPR 2025 | Rate-distortion optimized | Learned codec with ECVQ | | EAGLES | Quantized embeddings | ECCV 2024 | 8–16 bit | Coarse-to-fine training + pruning | | TC-GS | Tri-plane representation | IEEE 2026 | Implicit via tri-plane | Replaces per-Gaussian SH with shared tri-plane |

### Bit-width Selection Guide

| Attribute | 4–5 bit | 6–8 bit | 8–12 bit | 12–16 bit | |-----------|---------|---------|----------|-----------| | Position | Edge only — visible artifacts | Mobile/Web acceptable | Desktop recommended | Lossless-range | | SH (dc) | Not recommended | Edge/mobile | Desktop | High-fidelity | | SH (rest) | Aggressive mobile | Mobile/Web | Desktop | Unnecessary | | Opacity | Acceptable (post-sigmoid) | Recommended | Overkill | Overkill | | Scale | Log-space 4-bit risky | Log-space 6–8 bit | Recommended | Overkill | | Rotation | 8-bit often sufficient | Standard | Unnecessary | Overkill |

**Rule of thumb:** Position and SH dominate quality; allocate more bits there. Opacity and rotation tolerate aggressive quantization.

## Step 4: Pruning

### Strategies

| Strategy | Method | Venue | Compression | Quality Impact | |----------|--------|-------|-------------|----------------| | **Coreset-based** | Provable Pruning via Coresets | arXiv 2026 | Theoretical guarantee | Minimal — multiplicative approximation | | **Bayesian** | DP-Splat | arXiv 2026 | Automatic complexity control | Minimal — DP prior converges to optimal count | | **Training-free semantic** | CoSAG | arXiv 2026 | 37–76× over LangSplatV2 | Minimal — zero fine-tuning, leverages CLIP features | | **Importance-based** | Prune Wisely (DoG) | CVPR 2026 | 90% reduction | Minimal — DoG avoids false positives | | **Variational** | VEDAL | arXiv 2026（venue 待核实） | 5.2x (0.31 dB drop) | Low — uncertainty-gated async pruning | | **Merge-based** | NanoGS | arXiv 2026 | Training-free | Mass-preserving moment matching | | **Global+Local** | LightGaussian | NeurIPS 2024 | 15x | SVD distillation compensates | | **Render-aware** | GETA-3DGS | arXiv 2026 | ~5x storage | Transmittance-weighted saliency | | **Frequency-aware** | FAD-GS | CVPR 2024 | Frequency-separated | Separates low/high freq Gaussians | | **Memory-bounded** | Gaussians on a Diet | arXiv 2026 | 80% peak memory | Iterative growth+pruning | | **Hybrid** | HybridGS | CVPR 2025 | Explicit+implicit | Neural coding recovers pruned info | | **Budget-controlled** | MGS (Matryoshka) | arXiv 2026 | Continuous LoD | Any prefix of ordered set is coherent |

### Pruning Decision Flow

``` Need theoretical guarantees? ├── Yes → Provable Pruning via Coresets └── No ├── Training-free requirement? │ ├── Yes → NanoGS (merge) or LightGaussian (post-training) │ └── No │ ├── Can retrain/fine-tune after pruning? │ │ ├── Yes → Prune Wisely (DoG) + finetune, or VEDAL │ │ └── No → NanoGS or GETA-3DGS (auto, no per-scene thresholds) │ └── Need continuous quality levels? │ └── Yes → MGS (Matryoshka stochastic budget training) ```

## Step 5: Vector Quantization

### VQ Pipeline

``` Gaussian Attributes │ ▼ [1] Attribute Grouping ── group by type (position, SH, scale/rotation) │ ▼ [2] Codebook Learning ── K-means / learned / residual codebook │ ▼ [3] Assignment ── nearest-neighbor lookup per group │ ▼ [4] Residual Coding ── (optional) multi-level residual VQ │ ▼ [5] Entropy Coding ── arithmetic / ANS coding of indices │ ▼ Compressed Bitstream ```

### VQ Methods

| Method | Codebook Type | Venue | Compression | Key Feature | |--------|--------------|-------|-------------|-------------| | CompactGS | Learned per-attribute | ECCV 2024 | 10–15x | Simple codebook, minimal overhead | | VQGS | Residual codebook | CVPR 2025 | High-ratio | Multi-level residual improves quality | | RDO-Gaussian | ECVQ (entropy-constrained) | ECCV 2024 | 40x+ | Rate-distortion optimized VQ | | CAGS | VQ + LoD layers | SIGGRAPH 2026 | Adaptive | VQ establishes quality LoDs for streaming | | CGVQ | Clustered codebook | SIGGRAPH 2026 Poster | 20% bpp reduction | Cluster-guided grouping before quant | | Sp2403GS | Codebook + pruning | CVPR 2024 | Combined | Importance-based codebook selection | | HAC | Hash-grid context | ECCV 2024 | ~100x | Context modeling for entropy coding | | CompGS | Importance-aware | CVPR 2025 | Progressive | Progressive decoding support |

### Codebook Design Rules

1. **Codebook size K**: 256 (8-bit index) is standard; 1024 (10-bit) for quality; 64 (6-bit) for extreme compression 2. **Grouping strategy**: Group by attribute type (position separate from SH); within SH, separate DC from higher-order 3. **Residual levels**: 1 level = 10–20x; 2 levels = 20–50x; 3 levels = diminishing returns 4. **LoD integration**: Each residual level can serve as a LoD tier (see Step 6)

## Step 6: Streaming & LoD

### Static Scene Streaming

| Method | LoD Mechanism | Venue | Latency Reduction | Key Feature | |--------|--------------|-------|-------------------|-------------| | StreamLoD-GS | View-dependent LoD levels | arXiv 2026 | Progressive | Bandwidth-adaptive FVV delivery | | HGS | Hierarchical Gaussian structuring | CVPR 2025 | Progressive | Level-of-detail Gaussian hierarchy | | GS-Stream | Progressive chunk delivery | CVPR 2025 | Bandwidth-adaptive | Chunk-based 3DGS streaming | | EvoGS | Evolution Tree (wavelet-inspired) | arXiv 2026 | 2.4x payload reduction | Continuous parent-child refinement | | MGS | Stochastic budget training | arXiv 2026 | Continuous | Any prefix = coherent render | | SCube | VoxSplats + hierarchical LOD | NeurIPS 2024 | Hierarchical | Voxelized splat for large-scale | | CAGS | VQ-based LoD + reference image | SIGGRAPH 2026 | +5–20 dB PSNR | Server-side low-res reference corrects color |

### Dynamic (4DGS) Streaming

| Method | Mechanism | Venue | First-frame Latency | Key Feature | |--------|-----------|-------|---------------------|-------------|

--- name: 3dgs-compression-deploy description: "3DGS compression-to-deployment pipeline: quantization (scalar/VQ/mixed-precision), pruning (coreset/adaptive/variational/merge/Bayesian), progressive streaming & LoD, Web/WebGPU/mobile deployment, hardware acceleration (Tensor Core/GEMM/FPGA/ASIC), training-free semantic compression. Covers 53+ methods across 6 compression categories. Use when: compressing 3DGS models, deploying 3DGS to web/mobile/edge, selecting quantization bit-width, designing streaming pipelines, 3DGS压缩/量化/剪枝/部署/流式传输/移动端/硬件加速." license: Apache-2.0 user-invocable: true metadata: version: "1.1.1" author: jaccen tags: ["3dgs", "gaussian-splatting", "compression", "quantization", "pruning", "vector-quantization", "streaming", "deployment", "mobile", "webgpu", "tensor-core", "hardware-acceleration", "bayesian", "semantic-compression"] when_to_use: - "Compress a 3DGS model for storage or transmission" - "Deploy 3DGS to web browser, mobile device, or edge hardware" - "Select quantization scheme, bit-width, or pruning strategy" - "Design progressive streaming or Level-of-Detail pipeline" - "Evaluate hardware acceleration options (Tensor Core, GEMM, FPGA, ASIC)" - "3DGS压缩 / 量化 / 剪枝 / 部署 / 流式传输 / 移动端 / 硬件加速" ---

# 3DGS Compression & Deployment

> End-to-end pipeline from raw 3DGS model to deployed application. Covers 6 compression categories + 4 deployment targets + hardware acceleration.

## Capabilities

- Analyze 3DGS model attributes (position, SH, opacity, scale, rotation) and recommend compression strategy - Select quantization method and bit-width per attribute (scalar, VQ, mixed-precision) - Design pruning pipeline (coreset, adaptive, variational, merge-based) - Plan VQ codebook architecture and residual coding - Architect progressive streaming and LoD systems (static and 4D dynamic) - Guide platform-specific deployment (WebGL, WebGPU, iOS/Android, desktop) - Evaluate hardware acceleration paths (Tensor Core, GEMM, FPGA, ASIC) - Estimate compression ratio, quality loss, and rendering speed for each method combination

## Compression Pipeline

``` Raw 3DGS Model │ ▼ [Step 1] Analysis ── attribute profiling, bottleneck identification │ ▼ [Step 2] Strategy Selection ── target platform → compression recipe │ ▼ [Step 3] Pruning ── reduce Gaussian count (coreset / adaptive / variational / merge) │ ▼ [Step 4] Quantization ── reduce per-attribute bit-width (scalar / VQ / mixed-precision) │ ▼ [Step 5] Vector Quantization ── codebook-based attribute compression (optional, replaces/augments Step 4) │ ▼ [Step 6] Streaming & LoD ── progressive loading structure for network delivery │ ▼ [Step 7] Deployment ── platform-specific renderer and runtime │ ▼ Deployed Application (Web / Mobile / Desktop / Edge) ```

## Step 1: Analysis

Profile the 3DGS model before selecting compression methods:

| Attribute | FP32 Size | Typical Range | Sensitivity to Quantization | |-----------|-----------|---------------|---------------------------| | Position (μ) | 12B/Gaussian | Scene bounds | High — direct geometry impact | | SH (degree 0–3) | 48B/Gaussian | [-1, 1] per coeff | Medium-High — visual quality driver | | Opacity (α) | 4B/Gaussian | [0, 1] | Medium — pruning signal | | Scale (s) | 12B/Gaussian | [1e-5, 1e2] | Medium — anisotropy sensitive | | Rotation (q) | 16B/Gaussian | Unit quaternion | Low-Medium — can tolerate 8-bit |

**Profiling checklist:** 1. Total Gaussian count N and file size S 2. Target platform constraints (memory budget, bandwidth, GPU capability) 3. Quality floor (minimum acceptable PSNR/SSIM) 4. Required FPS threshold 5. Whether dynamic (4DGS) or static scene

## Step 2: Compression Strategy Selection

Decision tree by target platform:

``` Target Platform? ├── Web (WebGL/WebGPU) │ ├── Bandwidth-limited → Pruning + VQ + Streaming (CAGS/HGS pipeline) │ └── Compute-limited → Aggressive pruning + low SH degree + Flux-GS ├── Mobile (iOS/Android) │ ├── Real-time required → Mobile-GS pipeline (depth-aware OIT + distillation + pruning) │ └── Quality priority → MesonGS++ (mixed-precision, budget-controlled) + NanoGS merge ├── Desktop (GPU ≥ RTX 3060) │ ├── Max quality → Light quantization only (8-10 bit, ContextGS entropy coding) │ └── Large scene → Pruning + Streaming + HiGS hierarchical tiles └── Edge / Embedded ├── FPGA targeted → SpqGS (hardware-friendly quantization) + Axis-Shared Accelerator └── Low-power GPU → VEDAL pruning + 4-6 bit quantization + PocketGS on-device ```

**Combined target table:**

| Target | Typical Gaussian Budget | Bit-width Range | Streaming | Key Methods | |--------|------------------------|-----------------|-----------|-------------| | Web | 100K–500K | 4–8 bit | Required | CAGS, StreamLoD-GS, Spark 2.0 | | Mobile | 50K–200K | 4–8 bit | Optional | Mobile-GS, Flux-GS, PocketGS | | Desktop | 500K–5M | 8–16 bit | For large scenes | MesonGS++, HiGS, gsplat | | Edge/FPGA | 10K–100K | 2–6 bit | Required | SpqGS, VEDAL, GEMM-GS |

## Step 3: Quantization

### Method Selection

| Method | Type | Venue | Bit-width | Key Feature | |--------|------|-------|-----------|-------------| | MesonGS++ | Mixed-precision (post-training) | arXiv 2026 | 4–16 bit per attribute | 0-1 ILP hyperparameter search, 34x compression | | GETA-3DGS | Joint pruning + quantization | arXiv 2026 | 4–8 bit heterogeneous | Render-aware saliency, QADG dependency graph | | GSQ | Learned step size | CVPR 2025 | 4–8 bit | Group-wise quantization with learnable step | | ContextGS | Context-model entropy coding | NeurIPS 2024 | 8–16 bit | Anchor-level context replaces uniform quant | | SpqGS | Scalable parallel | CVPR 2025 | Hardware-friendly | Parallel bit allocation for FPGA/ASIC | | SOG-GS | Channel-grouped | CVPR 2025 | Per-channel | Preserves inter-Gaussian correlations | | ZipGS | Pruning + quant + entropy | CVPR 2025 | Variable | Volumetric entropy coding | | GaussianCodec | Entropy-constrained | CVPR 2025 | Rate-distortion optimized | Learned codec with ECVQ | | EAGLES | Quantized embeddings | ECCV 2024 | 8–16 bit | Coarse-to-fine training + pruning | | TC-GS | Tri-plane representation | IEEE 2026 | Implicit via tri-plane | Replaces per-Gaussian SH with shared tri-plane |

### Bit-width Selection Guide

| Attribute | 4–5 bit | 6–8 bit | 8–12 bit | 12–16 bit | |-----------|---------|---------|----------|-----------| | Position | Edge only — visible artifacts | Mobile/Web acceptable | Desktop recommended | Lossless-range | | SH (dc) | Not recommended | Edge/mobile | Desktop | High-fidelity | | SH (rest) | Aggressive mobile | Mobile/Web | Desktop | Unnecessary | | Opacity | Acceptable (post-sigmoid) | Recommended | Overkill | Overkill | | Scale | Log-space 4-bit risky | Log-space 6–8 bit | Recommended | Overkill | | Rotation | 8-bit often sufficient | Standard | Unnecessary | Overkill |

**Rule of thumb:** Position and SH dominate quality; allocate more bits there. Opacity and rotation tolerate aggressive quantization.

## Step 4: Pruning

### Strategies

| Strategy | Method | Venue | Compression | Quality Impact | |----------|--------|-------|-------------|----------------| | **Coreset-based** | Provable Pruning via Coresets | arXiv 2026 | Theoretical guarantee | Minimal — multiplicative approximation | | **Bayesian** | DP-Splat | arXiv 2026 | Automatic complexity control | Minimal — DP prior converges to optimal count | | **Training-free semantic** | CoSAG | arXiv 2026 | 37–76× over LangSplatV2 | Minimal — zero fine-tuning, leverages CLIP features | | **Importance-based** | Prune Wisely (DoG) | CVPR 2026 | 90% reduction | Minimal — DoG avoids false positives | | **Variational** | VEDAL | arXiv 2026（venue 待核实） | 5.2x (0.31 dB drop) | Low — uncertainty-gated async pruning | | **Merge-based** | NanoGS | arXiv 2026 | Training-free | Mass-preserving moment matching | | **Global+Local** | LightGaussian | NeurIPS 2024 | 15x | SVD distillation compensates | | **Render-aware** | GETA-3DGS | arXiv 2026 | ~5x storage | Transmittance-weighted saliency | | **Frequency-aware** | FAD-GS | CVPR 2024 | Frequency-separated | Separates low/high freq Gaussians | | **Memory-bounded** | Gaussians on a Diet | arXiv 2026 | 80% peak memory | Iterative growth+pruning | | **Hybrid** | HybridGS | CVPR 2025 | Explicit+implicit | Neural coding recovers pruned info | | **Budget-controlled** | MGS (Matryoshka) | arXiv 2026 | Continuous LoD | Any prefix of ordered set is coherent |

### Pruning Decision Flow

``` Need theoretical guarantees? ├── Yes → Provable Pruning via Coresets └── No ├── Training-free requirement? │ ├── Yes → NanoGS (merge) or LightGaussian (post-training) │ └── No │ ├── Can retrain/fine-tune after pruning? │ │ ├── Yes → Prune Wisely (DoG) + finetune, or VEDAL │ │ └── No → NanoGS or GETA-3DGS (auto, no per-scene thresholds) │ └── Need continuous quality levels? │ └── Yes → MGS (Matryoshka stochastic budget training) ```

## Step 5: Vector Quantization

### VQ Pipeline

``` Gaussian Attributes │ ▼ [1] Attribute Grouping ── group by type (position, SH, scale/rotation) │ ▼ [2] Codebook Learning ── K-means / learned / residual codebook │ ▼ [3] Assignment ── nearest-neighbor lookup per group │ ▼ [4] Residual Coding ── (optional) multi-level residual VQ │ ▼ [5] Entropy Coding ── arithmetic / ANS coding of indices │ ▼ Compressed Bitstream ```

### VQ Methods

| Method | Codebook Type | Venue | Compression | Key Feature | |--------|--------------|-------|-------------|-------------| | CompactGS | Learned per-attribute | ECCV 2024 | 10–15x | Simple codebook, minimal overhead | | VQGS | Residual codebook | CVPR 2025 | High-ratio | Multi-level residual improves quality | | RDO-Gaussian | ECVQ (entropy-constrained) | ECCV 2024 | 40x+ | Rate-distortion optimized VQ | | CAGS | VQ + LoD layers | SIGGRAPH 2026 | Adaptive | VQ establishes quality LoDs for streaming | | CGVQ | Clustered codebook | SIGGRAPH 2026 Poster | 20% bpp reduction | Cluster-guided grouping before quant | | Sp2403GS | Codebook + pruning | CVPR 2024 | Combined | Importance-based codebook selection | | HAC | Hash-grid context | ECCV 2024 | ~100x | Context modeling for entropy coding | | CompGS | Importance-aware | CVPR 2025 | Progressive | Progressive decoding support |

### Codebook Design Rules

1. **Codebook size K**: 256 (8-bit index) is standard; 1024 (10-bit) for quality; 64 (6-bit) for extreme compression 2. **Grouping strategy**: Group by attribute type (position separate from SH); within SH, separate DC from higher-order 3. **Residual levels**: 1 level = 10–20x; 2 levels = 20–50x; 3 levels = diminishing returns 4. **LoD integration**: Each residual level can serve as a LoD tier (see Step 6)

## Step 6: Streaming & LoD

### Static Scene Streaming

| Method | LoD Mechanism | Venue | Latency Reduction | Key Feature | |--------|--------------|-------|-------------------|-------------| | StreamLoD-GS | View-dependent LoD levels | arXiv 2026 | Progressive | Bandwidth-adaptive FVV delivery | | HGS | Hierarchical Gaussian structuring | CVPR 2025 | Progressive | Level-of-detail Gaussian hierarchy | | GS-Stream | Progressive chunk delivery | CVPR 2025 | Bandwidth-adaptive | Chunk-based 3DGS streaming | | EvoGS | Evolution Tree (wavelet-inspired) | arXiv 2026 | 2.4x payload reduction | Continuous parent-child refinement | | MGS | Stochastic budget training | arXiv 2026 | Continuous | Any prefix = coherent render | | SCube | VoxSplats + hierarchical LOD | NeurIPS 2024 | Hierarchical | Voxelized splat for large-scale | | CAGS | VQ-based LoD + reference image | SIGGRAPH 2026 | +5–20 dB PSNR | Server-side low-res reference corrects color |

### Dynamic (4DGS) Streaming

| Method | Mechanism | Venue | First-frame Latency | Key Feature | |--------|-----------|-------|---------------------|-------------|

--- name: 3dgs-compression-deploy description: "3DGS compression-to-deployment pipeline: quantization (scalar/VQ/mixed-precision), pruning (coreset/adaptive/variational/merge/Bayesian), progressive streaming & LoD, Web/WebGPU/mobile deployment, hardware acceleration (Tensor Core/GEMM/FPGA/ASIC), training-free semantic compression. Covers 53+ methods across 6 compression categories. Use when: compressing 3DGS models, deploying 3DGS to web/mobile/edge, selecting quantization bit-width, designing streaming pipelines, 3DGS压缩/量化/剪枝/部署/流式传输/移动端/硬件加速." license: Apache-2.0 user-invocable: true metadata: version: "1.1.1" author: jaccen tags: ["3dgs", "gaussian-splatting", "compression", "quantization", "pruning", "vector-quantization", "streaming", "deployment", "mobile", "webgpu", "tensor-core", "hardware-acceleration", "bayesian", "semantic-compression"] when_to_use: - "Compress a 3DGS model for storage or transmission" - "Deploy 3DGS to web browser, mobile device, or edge hardware" - "Select quantization scheme, bit-width, or pruning strategy" - "Design progressive streaming or Level-of-Detail pipeline" - "Evaluate hardware acceleration options (Tensor Core, GEMM, FPGA, ASIC)" - "3DGS压缩 / 量化 / 剪枝 / 部署 / 流式传输 / 移动端 / 硬件加速" ---

# 3DGS Compression & Deployment

> End-to-end pipeline from raw 3DGS model to deployed application. Covers 6 compression categories + 4 deployment targets + hardware acceleration.

## Capabilities

- Analyze 3DGS model attributes (position, SH, opacity, scale, rotation) and recommend compression strategy - Select quantization method and bit-width per attribute (scalar, VQ, mixed-precision) - Design pruning pipeline (coreset, adaptive, variational, merge-based) - Plan VQ codebook architecture and residual coding - Architect progressive streaming and LoD systems (static and 4D dynamic) - Guide platform-specific deployment (WebGL, WebGPU, iOS/Android, desktop) - Evaluate hardware acceleration paths (Tensor Core, GEMM, FPGA, ASIC) - Estimate compression ratio, quality loss, and rendering speed for each method combination

## Compression Pipeline

``` Raw 3DGS Model │ ▼ [Step 1] Analysis ── attribute profiling, bottleneck identification │ ▼ [Step 2] Strategy Selection ── target platform → compression recipe │ ▼ [Step 3] Pruning ── reduce Gaussian count (coreset / adaptive / variational / merge) │ ▼ [Step 4] Quantization ── reduce per-attribute bit-width (scalar / VQ / mixed-precision) │ ▼ [Step 5] Vector Quantization ── codebook-based attribute compression (optional, replaces/augments Step 4) │ ▼ [Step 6] Streaming & LoD ── progressive loading structure for network delivery │ ▼ [Step 7] Deployment ── platform-specific renderer and runtime │ ▼ Deployed Application (Web / Mobile / Desktop / Edge) ```

## Step 1: Analysis

Profile the 3DGS model before selecting compression methods:

| Attribute | FP32 Size | Typical Range | Sensitivity to Quantization | |-----------|-----------|---------------|---------------------------| | Position (μ) | 12B/Gaussian | Scene bounds | High — direct geometry impact | | SH (degree 0–3) | 48B/Gaussian | [-1, 1] per coeff | Medium-High — visual quality driver | | Opacity (α) | 4B/Gaussian | [0, 1] | Medium — pruning signal | | Scale (s) | 12B/Gaussian | [1e-5, 1e2] | Medium — anisotropy sensitive | | Rotation (q) | 16B/Gaussian | Unit quaternion | Low-Medium — can tolerate 8-bit |

**Profiling checklist:** 1. Total Gaussian count N and file size S 2. Target platform constraints (memory budget, bandwidth, GPU capability) 3. Quality floor (minimum acceptable PSNR/SSIM) 4. Required FPS threshold 5. Whether dynamic (4DGS) or static scene

## Step 2: Compression Strategy Selection

Decision tree by target platform:

``` Target Platform? ├── Web (WebGL/WebGPU) │ ├── Bandwidth-limited → Pruning + VQ + Streaming (CAGS/HGS pipeline) │ └── Compute-limited → Aggressive pruning + low SH degree + Flux-GS ├── Mobile (iOS/Android) │ ├── Real-time required → Mobile-GS pipeline (depth-aware OIT + distillation + pruning) │ └── Quality priority → MesonGS++ (mixed-precision, budget-controlled) + NanoGS merge ├── Desktop (GPU ≥ RTX 3060) │ ├── Max quality → Light quantization only (8-10 bit, ContextGS entropy coding) │ └── Large scene → Pruning + Streaming + HiGS hierarchical tiles └── Edge / Embedded ├── FPGA targeted → SpqGS (hardware-friendly quantization) + Axis-Shared Accelerator └── Low-power GPU → VEDAL pruning + 4-6 bit quantization + PocketGS on-device ```

**Combined target table:**

| Target | Typical Gaussian Budget | Bit-width Range | Streaming | Key Methods | |--------|------------------------|-----------------|-----------|-------------| | Web | 100K–500K | 4–8 bit | Required | CAGS, StreamLoD-GS, Spark 2.0 | | Mobile | 50K–200K | 4–8 bit | Optional | Mobile-GS, Flux-GS, PocketGS | | Desktop | 500K–5M | 8–16 bit | For large scenes | MesonGS++, HiGS, gsplat | | Edge/FPGA | 10K–100K | 2–6 bit | Required | SpqGS, VEDAL, GEMM-GS |

## Step 3: Quantization

### Method Selection

| Method | Type | Venue | Bit-width | Key Feature | |--------|------|-------|-----------|-------------| | MesonGS++ | Mixed-precision (post-training) | arXiv 2026 | 4–16 bit per attribute | 0-1 ILP hyperparameter search, 34x compression | | GETA-3DGS | Joint pruning + quantization | arXiv 2026 | 4–8 bit heterogeneous | Render-aware saliency, QADG dependency graph | | GSQ | Learned step size | CVPR 2025 | 4–8 bit | Group-wise quantization with learnable step | | ContextGS | Context-model entropy coding | NeurIPS 2024 | 8–16 bit | Anchor-level context replaces uniform quant | | SpqGS | Scalable parallel | CVPR 2025 | Hardware-friendly | Parallel bit allocation for FPGA/ASIC | | SOG-GS | Channel-grouped | CVPR 2025 | Per-channel | Preserves inter-Gaussian correlations | | ZipGS | Pruning + quant + entropy | CVPR 2025 | Variable | Volumetric entropy coding | | GaussianCodec | Entropy-constrained | CVPR 2025 | Rate-distortion optimized | Learned codec with ECVQ | | EAGLES | Quantized embeddings | ECCV 2024 | 8–16 bit | Coarse-to-fine training + pruning | | TC-GS | Tri-plane representation | IEEE 2026 | Implicit via tri-plane | Replaces per-Gaussian SH with shared tri-plane |

### Bit-width Selection Guide

| Attribute | 4–5 bit | 6–8 bit | 8–12 bit | 12–16 bit | |-----------|---------|---------|----------|-----------| | Position | Edge only — visible artifacts | Mobile/Web acceptable | Desktop recommended | Lossless-range | | SH (dc) | Not recommended | Edge/mobile | Desktop | High-fidelity | | SH (rest) | Aggressive mobile | Mobile/Web | Desktop | Unnecessary | | Opacity | Acceptable (post-sigmoid) | Recommended | Overkill | Overkill | | Scale | Log-space 4-bit risky | Log-space 6–8 bit | Recommended | Overkill | | Rotation | 8-bit often sufficient | Standard | Unnecessary | Overkill |

**Rule of thumb:** Position and SH dominate quality; allocate more bits there. Opacity and rotation tolerate aggressive quantization.

## Step 4: Pruning

### Strategies

| Strategy | Method | Venue | Compression | Quality Impact | |----------|--------|-------|-------------|----------------| | **Coreset-based** | Provable Pruning via Coresets | arXiv 2026 | Theoretical guarantee | Minimal — multiplicative approximation | | **Bayesian** | DP-Splat | arXiv 2026 | Automatic complexity control | Minimal — DP prior converges to optimal count | | **Training-free semantic** | CoSAG | arXiv 2026 | 37–76× over LangSplatV2 | Minimal — zero fine-tuning, leverages CLIP features | | **Importance-based** | Prune Wisely (DoG) | CVPR 2026 | 90% reduction | Minimal — DoG avoids false positives | | **Variational** | VEDAL | arXiv 2026（venue 待核实） | 5.2x (0.31 dB drop) | Low — uncertainty-gated async pruning | | **Merge-based** | NanoGS | arXiv 2026 | Training-free | Mass-preserving moment matching | | **Global+Local** | LightGaussian | NeurIPS 2024 | 15x | SVD distillation compensates | | **Render-aware** | GETA-3DGS | arXiv 2026 | ~5x storage | Transmittance-weighted saliency | | **Frequency-aware** | FAD-GS | CVPR 2024 | Frequency-separated | Separates low/high freq Gaussians | | **Memory-bounded** | Gaussians on a Diet | arXiv 2026 | 80% peak memory | Iterative growth+pruning | | **Hybrid** | HybridGS | CVPR 2025 | Explicit+implicit | Neural coding recovers pruned info | | **Budget-controlled** | MGS (Matryoshka) | arXiv 2026 | Continuous LoD | Any prefix of ordered set is coherent |

### Pruning Decision Flow

``` Need theoretical guarantees? ├── Yes → Provable Pruning via Coresets └── No ├── Training-free requirement? │ ├── Yes → NanoGS (merge) or LightGaussian (post-training) │ └── No │ ├── Can retrain/fine-tune after pruning? │ │ ├── Yes → Prune Wisely (DoG) + finetune, or VEDAL │ │ └── No → NanoGS or GETA-3DGS (auto, no per-scene thresholds) │ └── Need continuous quality levels? │ └── Yes → MGS (Matryoshka stochastic budget training) ```

## Step 5: Vector Quantization

### VQ Pipeline

``` Gaussian Attributes │ ▼ [1] Attribute Grouping ── group by type (position, SH, scale/rotation) │ ▼ [2] Codebook Learning ── K-means / learned / residual codebook │ ▼ [3] Assignment ── nearest-neighbor lookup per group │ ▼ [4] Residual Coding ── (optional) multi-level residual VQ │ ▼ [5] Entropy Coding ── arithmetic / ANS coding of indices │ ▼ Compressed Bitstream ```

### VQ Methods

| Method | Codebook Type | Venue | Compression | Key Feature | |--------|--------------|-------|-------------|-------------| | CompactGS | Learned per-attribute | ECCV 2024 | 10–15x | Simple codebook, minimal overhead | | VQGS | Residual codebook | CVPR 2025 | High-ratio | Multi-level residual improves quality | | RDO-Gaussian | ECVQ (entropy-constrained) | ECCV 2024 | 40x+ | Rate-distortion optimized VQ | | CAGS | VQ + LoD layers | SIGGRAPH 2026 | Adaptive | VQ establishes quality LoDs for streaming | | CGVQ | Clustered codebook | SIGGRAPH 2026 Poster | 20% bpp reduction | Cluster-guided grouping before quant | | Sp2403GS | Codebook + pruning | CVPR 2024 | Combined | Importance-based codebook selection | | HAC | Hash-grid context | ECCV 2024 | ~100x | Context modeling for entropy coding | | CompGS | Importance-aware | CVPR 2025 | Progressive | Progressive decoding support |

### Codebook Design Rules

1. **Codebook size K**: 256 (8-bit index) is standard; 1024 (10-bit) for quality; 64 (6-bit) for extreme compression 2. **Grouping strategy**: Group by attribute type (position separate from SH); within SH, separate DC from higher-order 3. **Residual levels**: 1 level = 10–20x; 2 levels = 20–50x; 3 levels = diminishing returns 4. **LoD integration**: Each residual level can serve as a LoD tier (see Step 6)

## Step 6: Streaming & LoD

### Static Scene Streaming

| Method | LoD Mechanism | Venue | Latency Reduction | Key Feature | |--------|--------------|-------|-------------------|-------------| | StreamLoD-GS | View-dependent LoD levels | arXiv 2026 | Progressive | Bandwidth-adaptive FVV delivery | | HGS | Hierarchical Gaussian structuring | CVPR 2025 | Progressive | Level-of-detail Gaussian hierarchy | | GS-Stream | Progressive chunk delivery | CVPR 2025 | Bandwidth-adaptive | Chunk-based 3DGS streaming | | EvoGS | Evolution Tree (wavelet-inspired) | arXiv 2026 | 2.4x payload reduction | Continuous parent-child refinement | | MGS | Stochastic budget training | arXiv 2026 | Continuous | Any prefix = coherent render | | SCube | VoxSplats + hierarchical LOD | NeurIPS 2024 | Hierarchical | Voxelized splat for large-scale | | CAGS | VQ-based LoD + reference image | SIGGRAPH 2026 | +5–20 dB PSNR | Server-side low-res reference corrects color |

### Dynamic (4DGS) Streaming

| Method | Mechanism | Venue | First-frame Latency | Key Feature | |--------|-----------|-------|---------------------|-------------|