Research map / Image generation / Image restoration
Practical Full Resolution Learned Lossless Image 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 L3C instead of FLIF | Lossless image compression · Open Images | bits per sub-pixel: 2.867 -> 2.991 | Hurt |
| Use L3C instead of FLIF | Lossless image compression · DIV2K | bits per sub-pixel: 2.911 -> 3.094 | Hurt |
| Use L3C instead of FLIF | Lossless image compression · RAISE-1k | bits per sub-pixel: 2.084 -> 2.387 | Hurt |
| Use L3C instead of PixelCNN | ImageNet32 lossless compression · ImageNet32 | bits per sub-pixel: 3.83 -> 4.76 | Hurt |
| Use L3C instead of PixelCNN | ImageNet64 lossless compression · ImageNet64 | bits per sub-pixel: 3.57 -> 4.42 | Hurt |
| Learn feature representations rather than use a bicubic RGB pyramid (ablation) | RGB baseline · Open Images, DIV2K, and RAISE-1k | bits per sub-pixel: 3.298 -> 2.991; 3.418 -> 3.094; 2.572 -> 2.387 | Helped |
| Train a separate predictor for each scale instead of sharing one predictor (ablation) | RGB Shared baseline · Open Images, DIV2K, and RAISE-1k | bits per sub-pixel: 4.314 -> 3.298; 4.429 -> 3.418; 3.779 -> 2.572 | Helped |
| Add BatchNorm (ablation) | L3C · Image compression | bits per sub-pixel | Hurt |
| Replace stacked atrous convolutions with a single convolution (ablation) | L3C · Image compression | bits per sub-pixel | Hurt |
| Stop gradients from propagating through loss targets (ablation) | L3C · Learned feature representations | cross-entropy | Hurt |
| Use a smoother soft quantization setting, with σq=2 (ablation) | L3C · Image compression training | training performance | Helped |
| Increase quantization levels L rather than feature channels C (ablation) | L3C · Image compression training | training performance | Helped |
Closest papers by meaning
- Learning Better Lossless Compression Using Lossy Compression (CVPR 2020 · similarity 0.91)
- Efficient Learned Image Compression without Entropy Coding (ICML 2026 · similarity 0.90)
- Towards Efficient Image Compression Without Autoregressive Models (NeurIPS 2023 · similarity 0.90)
- Large Language Models for Lossless Image Compression: Next-Pixel Prediction in Language Space is All You Need (NeurIPS 2025 · similarity 0.88)
- GLIC: General Format Learned Image Compression (AAAI 2025 · similarity 0.88)
- CALLIC: Content Adaptive Learning for Lossless Image Compression (AAAI 2025 · similarity 0.87)
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.