Research map / Architectures and efficiency / RNNs and normalization
SHACIRA: Scalable HAsh-grid Compression for Implicit Neural Representations
Sharath Girish, Abhinav Shrivastava, Kamal Gupta
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 |
|---|---|---|---|
| reparameterize feature grids with quantized latents and entropy regularization on large images | Instant-NGP · Cosmic-Cliffs image | BPP: 0.63 -> 0.11 BPP at 38.78 -> 38.79 PSNR | Improved |
| compress Instant-NGP feature grid with SHACIRA on RTMV NeRF scenes | Instant-NGP · RTMV 10 brick scenes | Storage: 48.9MB -> 1.03MB; PSNR 31.88 -> 32.14 | Improved |
| add entropy regularization on latents (ablation) | SHACIRA · UVG-F Jockey image | PSNR-BPP trade-off | Improved |
| increase entropy regularization weight to 4.0e-4 (ablation) | SHACIRA · UVG-F Jockey image | PSNR-BPP trade-off | Hurt |
| add annealing of discrete latents instead of STE alone (ablation) | SHACIRA · UVG-F Jockey image | PSNR-BPP trade-off | Improved |
| use feature-grid SHACIRA instead of MLP-based positional INR | Positional INR · UVG-F | PSNR: 33.17 -> 37.74 | Improved |
| use SHACIRA instead of NeRV for video | UVG video · UVG video | PSNR: 35.54 -> 35.01 PSNR; 0.66 -> 0.21 BPP | Mixed |
| use SHACIRA feature grid vs MLP-based INR for convergence speed | Kodak image · Kodak | encoding time: 600s vs 53s at similar PSNR and BPP | Improved |
| use SHACIRA at very high resolution (Pearl, Tokyo) | Instant-NGP · Pearl and Tokyo | PSNR: 29.62 -> 29.44; 31.87 -> 31.62 | Mixed |
| use SHACIRA compared with VQAD | VQAD · RTMV NeRF | Storage: 0.55MB -> 0.43MB; PSNR 31.45 -> 31.46 | Improved |
Closest papers by meaning
- SINR: Sparsity Driven Compressed Implicit Neural Representations (CVPR 2025 · similarity 0.87)
- NVRC: Neural Video Representation Compression (NeurIPS 2024 · similarity 0.86)
- Grids Often Outperform Implicit Neural Representation at Compressing Dense Signals (NeurIPS 2025 · similarity 0.85)
- Neural NeRF Compression (ICML 2024 · similarity 0.85)
- Combining Frame and GOP Embeddings for Neural Video Representation (CVPR 2024 · similarity 0.85)
- Boosting Neural Representations for Videos with a Conditional Decoder (CVPR 2024 · similarity 0.85)
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.
