Research map / Image generation / Image restoration
CoD-Lite: Real-Time Diffusion-Based Generative Image Compression
Zhaoyang Jia, Naifu Xue, Zihan Zheng, Jiahao Li, Bin Li, Xiaoyi Zhang, Zongyu Guo, Yuan Zhang, Houqiang Li, Yan Lu
This paper tried a common change: Train longer
train longer · Kodak 0.0312 bpp · FID: 32.8 -> 31.5 Improved
What 387 accepted papers report for this change
- 211 improved
- 62 no clear effect
- 30 mixed
- 84 hurt
- Lightweight Federated Incremental Learning via Decoupled Replay ICML 2026 · increase local training epochs from 5 to 40 · CIFAR10 (α=1.0) · test accuracyImproved
- When Do Hallucinations Arise? A Graph Perspective on the Evolution of Path Reuse and Path Compression ICML 2026 · train longer on the full path search space (overfitting stage) · SBM graph, 10 communities, pin=0.6 · accuracy: 1.0 -> 0.98Hurt
- Causes and Consequences of Representational Similarity in Machine Learning Models ICML 2026 · train longer on LLM continued pretraining to test insufficient adaptation · perplexity across overlap settings · perplexityImproved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 |
|---|---|---|---|
| unconditional generation-oriented diffusion pre-training for large codec (ablation) | 700M diffusion codec · ImageNet 256, 0.0156 bpp | FID: 23.7 -> 15.4 | Improved |
| unconditional generation-oriented diffusion pre-training for small codec (ablation) | 34M diffusion codec · ImageNet 256, 0.0156 bpp | FID: 2.0 FID gain vs 8.3 for large model | No clear effect |
| class-conditioned pre-training instead of unconditional for small model (ablation) | 34M diffusion codec · ImageNet 256, 0.0156 bpp | FID | Improved |
| class-conditioned pre-training instead of unconditional for large model (ablation) | 700M diffusion codec · ImageNet 256, 0.0156 bpp | FID | Hurt |
| use compression-oriented CoD pre-training instead of generation-oriented (ablation) | 34M diffusion codec · ImageNet 256, 0.0156 bpp | FID: -3.5 FID (34M); -3.4 FID (700M) | Improved |
| add CoD pre-training to from-scratch baseline (ablation) | one-step codec · Kodak 0.0312 bpp | FID: 41.1 -> 29.6 | Improved |
| replace global attention with local window attention in CoD pre-training (ablation) | CoD · Kodak 0.0039 bpp pre-train | FID: 59.7 -> 60.2 | No clear effect |
| replace attention by depth-wise conv in CoD pre-training (ablation) | CoD · Kodak 0.0039 bpp pre-train | FID: 59.7 -> 62.5 | Hurt |
| replace attention by depth-wise conv in one-step codec (ablation) | one-step DiT codec · Kodak 0.0312 bpp | FID and decode speed: FID 37.5 -> 41.4; 331.2 ms -> 23.6 ms | Mixed |
| add DMD distillation loss from large CoD teacher (ablation) | conv one-step codec · Kodak 0.0312 bpp | FID: 41.4 -> 36.4 | Improved |
| add DMD distillation loss to DiT global attention codec (ablation) | DiT one-step codec · Kodak 0.0312 bpp | FID: 37.5 -> 35.4 | Improved |
| replace PatchGAN with projected GAN (DINOv2 features) (ablation) | CoD-Lite · Kodak 0.0312 bpp | FID: 36.4 -> 32.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- StableCodec — StableCodec: Taming One-Step Diffusion for Extreme Image Compression (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- One-Step Diffusion-Based Image Compression with Semantic Distillation (NeurIPS 2025 · similarity 0.91)
- Bridging the Gap between Gaussian Diffusion Models and Universal Quantization for Image Compression (CVPR 2025 · similarity 0.90)
- Progressive Compression with Universally Quantized Diffusion Models (ICLR 2025 · similarity 0.90)
- OSCAR: One-Step Diffusion Codec Across Multiple Bit-rates (NeurIPS 2025 · similarity 0.90)
- Lossy Compression with Pretrained Diffusion Models (ICLR 2025 · similarity 0.90)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.89)
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.
