Research map / Image generation / Image restoration
Learning Better Lossless Compression Using Lossy Compression
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 BPG to encode image content and a learned CNN to encode the residual | RC · Open Images; CLIC.mobile; CLIC.pro; DIV2K | bpsp: 2.790 vs 2.991; 2.538 vs 2.639; 2.933 vs 2.944; 3.079 vs 3.094 | Helped |
| Use the proposed BPG-plus-residual method rather than FLIF | RC · Open Images; CLIC.mobile; CLIC.pro; DIV2K | bpsp: 2.790 vs 2.867; 2.538 vs 2.492; 2.933 vs 2.784; 3.079 vs 2.911 | Mixed |
| Predict Q per image with the Q-Classifier instead of using fixed Q=14 (ablation) | RC · Open Images validation; DIV2K | bpsp: 2.801 -> 2.794; 3.096 -> 3.088 | Helped |
| Add per-image τ-optimization after Q-Classifier selection (ablation) | RC · Open Images validation; DIV2K | bpsp: 2.794 -> 2.790; 3.088 -> 3.079 | Helped |
| Use the per-image optimal Q' rather than fixed Q=14 (ablation) | RC · Open Images validation; DIV2K | bpsp: 2.801 -> 2.789; 3.096 -> 3.080 | Helped |
| Train with random Q values rather than one fixed Q (ablation) | Residual Compressor · Generalization performance | bpsp | Helped |
| Increase training crop size from 128×128 to 256×256 (ablation) | Residual Compressor · Open Images validation | bpsp: 2.854 -> 2.864 | Hurt |
| Increase training crop size from 128×128 to 512×512 (ablation) | Residual Compressor · Open Images validation | bpsp: 2.854 -> 2.877 | Hurt |
| Replace GDN layers with instance normalization (ablation) | Residual Compressor · Overall bitrates | bpsp | Hurt |
| Replace GDN layers with conditional instance normalization (ablation) | Residual Compressor · Overall bitrates | bpsp | Hurt |
Closest papers by meaning
- Practical Full Resolution Learned Lossless Image Compression (CVPR 2019 · similarity 0.91)
- Learning Scalable lY=-Constrained Near-Lossless Image Compression via Joint Lossy Image and Residual Compression (CVPR 2021 · similarity 0.90)
- Improved Lossy Image Compression With Priming and Spatially Adaptive Bit Rates for Recurrent Networks (CVPR 2018 · similarity 0.89)
- Lossy Compression for Lossless Prediction (NeurIPS 2021 · similarity 0.88)
- Lossless Compression with Probabilistic Circuits (ICLR 2022 · similarity 0.87)
- GLIC: General Format Learned Image Compression (AAAI 2025 · 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.