Research map / Image generation / Image restoration
PIC: Revisiting INR for Image Coding with Fast Encoding and Sub-Millisecond Decoding
Xiang Liu, Jinxiang Wang, Bin Chen, Zimo Liu, Mingyao Hong, Jiawei Li, Yaowei Wang, Shu-Tao Xia
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 |
|---|---|---|---|
| generate instance-specific synthesis weights via normalization and modulation instead of a shared synthesis net (ablation) | PIC · Kodak | RD performance | Improved |
| add run-length encoding of zeros after zigzag reorder (ablation) | PIC · Kodak | BD-rate: 150.68% -> 0% | Improved |
| extend RLE to symbols 1 or -1 as well (ablation) | PIC · Kodak | BD-rate: 0% -> 1.18% (0,1); 2.70% (-1,0,1) | Hurt |
| place instance-dependent MLP at the end of the synthesis net rather than start (ablation) | PIC · Kodak | BD-rate: 21.46% -> 0% | Improved |
| use synthesis width 16 instead of 8 or 32 (ablation) | PIC · Kodak | BD-rate: 3.03% (8) / 5.84% (32) -> 0% | Improved |
| width 8 padded to 16x16 on Tensor Cores (ablation) | PIC · Kodak | decoding speed | No clear effect |
| add modulation to Cool-Chic-like architecture (ablation) | N-O Cool-Chic* · Kodak | RD performance | Improved |
| feedforward INR codec versus per-image trained INR codecs | PIC vs Cool-Chic/COIN/GaussianImage · Kodak | encoding time: 6.99x10^5 ms -> 28.6 ms (vs GaussianImage) | Improved |
| PIC versus Cool-Chic v4.2 and Hyperprior on rate-distortion | PIC · Kodak | BD-rate vs JPEG: -12.78% vs -64.86% (Cool-Chic) and -57.58% (Hyperprior) | Hurt |
Closest papers by meaning
- NVRC: Neural Video Representation Compression (NeurIPS 2024 · similarity 0.87)
- PNVC: Towards Practical INR-based Video Compression (AAAI 2025 · similarity 0.87)
- SINR: Sparsity Driven Compressed Implicit Neural Representations (CVPR 2025 · similarity 0.87)
- Boosting Neural Representations for Videos with a Conditional Decoder (CVPR 2024 · similarity 0.87)
- Towards Lossless Implicit Neural Representation via Bit Plane Decomposition (CVPR 2025 · similarity 0.85)
- Implicit Representations via Operator Learning (ICML 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.
