Research map / Image generation / Image restoration
Lossy Image Compression with Conditional Diffusion Models
Read the paperCheck 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 |
|---|---|---|---|
| increase stochastic decoding noise gamma from 0 to 0.8 (ablation) | CDC X-prediction · Kodak example | LPIPS: 0.078 -> 0.054 | Helped |
| increase stochastic decoding noise gamma from 0 to 0.8 (ablation) | CDC X-prediction · Kodak example | MS-SSIM: 0.979 -> 0.973 | Hurt |
| increase gamma from 0.8 to 1 (ablation) | CDC X-prediction · Kodak example | LPIPS: 0.054 -> 0.058 | Hurt |
| use more decoding steps with stochastic decoding (ablation) | CDC X-prediction rho=0 · DIV2K | perceptual metrics | Helped |
| use more decoding steps with deterministic decoding (ablation) | CDC X-prediction rho=0 · DIV2K | distortion metrics | No clear effect |
| add LPIPS perceptual loss with rho=0.9 and stochastic decoding gamma=0.8 (ablation) | CDC X-prediction · DIV2K | perceptual metrics (FID etc.) | Helped |
| increase perceptual loss weight rho up to 0.9 (ablation) | CDC X-prediction · COCO | distortion metrics | Hurt |
| use X-prediction instead of epsilon-prediction with 17 decoding steps (ablation) | CDC · DIV2K | rate-quality tradeoff | Helped |
| replace Gaussian VAE decoder codecs with CDC rho=0 on PSNR | CDC rho=0 · DIV2K | PSNR | Hurt |
Closest papers by meaning
- Progressive Compression with Universally Quantized Diffusion Models (ICLR 2025 · similarity 0.91)
- OSCAR: One-Step Diffusion Codec Across Multiple Bit-rates (NeurIPS 2025 · similarity 0.91)
- Correcting Diffusion-Based Perceptual Image Compression with Privileged End-to-End Decoder (ICML 2024 · similarity 0.90)
- Compressed Image Generation with Denoising Diffusion Codebook Models (ICML 2025 · similarity 0.90)
- Lossy Compression with Pretrained Diffusion Models (ICLR 2025 · similarity 0.90)
- Bridging the Gap between Gaussian Diffusion Models and Universal Quantization for Image Compression (CVPR 2025 · similarity 0.90)
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.