Research map / Image generation / Image restoration
Idempotent Learned Image Compression with Right-Inverse
Yanghao Li, Tongda Xu, Yan Wang, Jingjing Liu, Ya-Qin Zhang
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 |
|---|---|---|---|
| use right-invertible blocked convolution framework instead of invertible flow codec [Helminger et al., 2021] | idempotent LIC · Kodak | BD-BR (%): 4.83 -> -28.75 | Improved |
| proposed idempotent vs Helminger | idempotent LIC · Kodak | BD-PSNR (dB): -0.21 -> 1.63 | Improved |
| proposed idempotent vs Helminger efficiency | idempotent LIC · Kodak 256x256x3 | GFLOPs: 15.89 -> 8.40 | Improved |
| proposed idempotent vs Helminger efficiency | idempotent LIC · Kodak 256x256x3 | enc-dec time (ms): 185 -> 110 | Improved |
| replace right-invertible blocked conv with invertible blocked conv (inv.) (ablation) | proposed idempotent codec · Kodak | RD performance (PSNR) | Hurt |
| replace blocked conv with invertible 1x1 convolution (w.conv1x1) (ablation) | proposed idempotent codec · Kodak | RD performance (PSNR) | Hurt |
| remove null-space enhancement at lower-bpp points (192 channels) (ablation) | proposed idempotent codec · Kodak | PSNR: 1 dB advantage of NE | Hurt |
| remove null-space enhancement at higher-bpp points (320 channels) (ablation) | proposed idempotent codec · Kodak | PSNR: smaller than 0.5 dB | Hurt |
| remove coupling enhancement (ablation) | proposed idempotent codec · Kodak | RD performance and block artifacts | Hurt |
| use first-type quantization (integer mean, w.int mean) instead of second-type (ablation) | proposed idempotent codec · Kodak | PSNR | Hurt |
| replace c-GDN with original GDN (non-idempotent) (ablation) | proposed idempotent codec · Kodak | RD performance | No clear effect |
| relax first layer to non-right-invertible (near-idempotent) vs Kim et al., 2020 | near-idempotent LIC · Kodak, 50 re-compressions | PSNR drop (dB): 8.26 -> 0.87 | Improved |
Closest papers by meaning
- Idempotence and Perceptual Image Compression (ICLR 2024 · similarity 0.87)
- On Disentangled Training for Nonlinear Transform in Learned Image Compression (ICLR 2025 · similarity 0.86)
- Few-Shot Domain Adaptation for Learned Image Compression (AAAI 2025 · similarity 0.86)
- Balanced Rate-Distortion Optimization in Learned Image Compression (CVPR 2025 · similarity 0.85)
- Towards Backward-Compatible Continual Learning of Image Compression (CVPR 2024 · similarity 0.85)
- Controllable Distortion-Perception Tradeoff Through Latent Diffusion for Neural Image Compression (AAAI 2025 · similarity 0.84)
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.
