Research map / Image generation / Image restoration
Multi-Realism Image Compression With a Conditional Generator
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 |
|---|---|---|---|
| Condition the decoder on receiver-selected realism weight β | Multi-realism compression model · MS-COCO 30K and CLIC 2020 | PSNR at high realism: ≈1dB | Helped |
| Use low β instead of non-generative compression | Multi-realism compression model · MS-COCO 30K, 0.26bpp | FID: about 40% | Helped |
| Increase β from 0 to 2.56 | Multi-realism compression model · Single representation | PSNR: 1.1dB | Hurt |
| Add the GAN loss (ablation) | Baseline -GAN -Charm · MS-COCO | FID | Helped |
| Increase GAN weight β beyond 2.56 (ablation) | Baseline +GAN -Charm · MS-COCO | FID | No clear effect |
| Set LPIPS weight C_P' to 4.26 (ablation) | Baseline +GAN -Charm · MS-COCO | FID: 4.26 | Helped |
| Replace Hyperprior entropy model with Charm (ablation) | GAN Baseline · MS-COCO | bitrate | Helped |
| Replace FourierCond with TableCond (ablation) | Multi-realism compression model · MS-COCO | distortion-realism trade-off | No clear effect |
| Add noise to the compressed representation before decoding (ablation) | GAN baseline · Rates studied in the paper | reconstruction quality | No clear effect |
| Train one decoder across sampled β values rather than for a single β | Multi-realism compression model · MS-COCO 30K | FID at β=0 | Helped |
Closest papers by meaning
- Towards image compression with perfect realism at ultra-low bitrates (ICLR 2024 · similarity 0.89)
- Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion (CVPR 2025 · similarity 0.87)
- Decouple Distortion from Perception: Region Adaptive Diffusion for Extreme-low Bitrate Perception Image Compression (CVPR 2025 · similarity 0.86)
- Compressed Image Generation with Denoising Diffusion Codebook Models (ICML 2025 · similarity 0.86)
- Generative Latent Coding for Ultra-Low Bitrate Image Compression (CVPR 2024 · similarity 0.86)
- Bridging the Gap between Gaussian Diffusion Models and Universal Quantization for Image Compression (CVPR 2025 · similarity 0.86)
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.