Research map / Probabilistic and scientific modeling / Variational and generative sampling
Compressing Images by Encoding Their Latent Representations with Relative Entropy Coding
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 |
|---|---|---|---|
| Use iREC instead of HiLLoC for single-image lossless compression | 24-level ResNet VAE compression · Cifar10 | bits per dimension: 24.51 -> 4.18 | Helped |
| Use iREC instead of HiLLoC for single-image lossless compression | 24-level ResNet VAE compression · ImageNet32 | bits per dimension: 26.80 -> 4.91 | Helped |
| Use iREC instead of HiLLoC for single-image lossless compression | 24-level ResNet VAE compression · Kodak | bits per dimension: 17.5 -> 3.67 | Helped |
| Use iREC instead of IDF | lossless image compression · Cifar10 | bits per dimension: 3.34 -> 4.18 | Hurt |
| Use iREC instead of IDF | lossless image compression · ImageNet32 | bits per dimension: 4.18 -> 4.91 | Hurt |
| Use iREC instead of FLIF | lossless image compression · Kodak | bits per dimension: 2.90 -> 3.67 | Hurt |
| Add beam search to auxiliary-variable importance sampling rather than using regular importance sampling (ablation) | iREC lossless compression on ImageNet32 · ImageNet32 validation | codelength overhead: 25-80% -> 15-25% | Helped |
| Increase beam width from B=5 to B=20 (ablation) | iREC lossless compression on ImageNet32 · ImageNet32 validation | codelength overhead: 15-25% -> 10-15% | Helped |
| Increase beam width beyond B=20 (ablation) | iREC lossless compression · ImageNet32 validation | codelength overhead | No clear effect |
| Anneal the KL divergence at the beginning of training (ablation) | lossy compression model · CLIC 2018 training | performance | No clear effect |
| Use iREC for lossy compression rather than quantized-latent methods | lossy image compression · Kodak | PSNR | Mixed |
Closest papers by meaning
- Improving Inference for Neural Image Compression (NeurIPS 2020 · similarity 0.91)
- Variational image compression with a scale hyperprior (ICLR 2018 · similarity 0.90)
- Practical lossless compression with latent variables using bits back coding (ICLR 2019 · similarity 0.90)
- Robustly overfitting latents for flexible neural image compression (NeurIPS 2024 · similarity 0.89)
- Learned Image Compression with Dictionary-based Entropy Model (CVPR 2025 · similarity 0.87)
- Generative Latent Coding for Ultra-Low Bitrate Image Compression (CVPR 2024 · 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.