Research map / 3D vision / Gaussian splatting
HAC: Hash-grid Assisted Context for 3D Gaussian Splatting Compression
Yihang Chen, Qianyi Wu, Weiyao Lin, Mehrtash Harandi, Jianfei Cai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add hash-grid context model for anchor attribute entropy coding | Scaffold-GS · Mip-NeRF360 | size (MB): 253.9 -> 15.26 (lowrate), PSNR 27.50 -> 27.53 | Improved |
| replace anchor features with hash features directly (ablation) | Scaffold-GS · Synthetic-NeRF | fidelity / training stability / FPS | Hurt |
| binarize anchor attributes instead of rounding with noise-based quantization (ablation) | HAC · anchor attributes | information preservation | Hurt |
| add mutual-information context from hash grid (vs all-zero hash grid) (ablation) | HAC · BungeeNeRF and Synthetic-NeRF | bit rate | Improved |
| add adaptive quantization module (AQM) (ablation) | HAC · BungeeNeRF and Synthetic-NeRF | fidelity | Improved |
| add adaptive offset masking (ablation) | HAC · BungeeNeRF and Synthetic-NeRF | bit rate | Improved |
| apply HAC instead of Morgenstern et al. dimension collapsing | 3DGS compression · Synthetic-NeRF | PSNR: 31.05 -> 33.24 (lowrate); size 2.20 -> 1.18 | Improved |
| increase anchor feature dimension to 50 and add entropy loss regularization | HAC · Mip-NeRF360 | PSNR: 27.50 -> 27.77 (highrate vs Scaffold-GS) | Improved |
| add HAC context models on training cost | Scaffold-GS · BungeeNeRF | training time (min): 15.1 -> 27.6 | Hurt |
| HAC pruning of invalid anchors on rendering speed | Scaffold-GS · BungeeNeRF | FPS: 232 -> 283 | Improved |
| HAC versus Scaffold-GS on Synthetic-NeRF rendering speed | Scaffold-GS · Synthetic-NeRF | FPS: 326 -> 341 (HAC); 3DGS 401 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Scaffold-GS — Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering (CVPR 2024) · better on 8 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ContextGS : Compact 3D Gaussian Splatting with Anchor Level Context Model (NeurIPS 2024 · similarity 0.93)
- Fast Feedforward 3D Gaussian Splatting Compression (ICLR 2025 · similarity 0.92)
- Optimized Minimal 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
- Compressed 3D Gaussian Splatting for Accelerated Novel View Synthesis (CVPR 2024 · similarity 0.91)
- Grid4D: 4D Decomposed Hash Encoding for High-Fidelity Dynamic Gaussian Splatting (NeurIPS 2024 · similarity 0.91)
- Locality-aware Gaussian Compression for Fast and High-quality Rendering (ICLR 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
