Research map / Image generation / Image restoration
Joint Autoregressive and Hierarchical Priors for Learned Image Compression
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 |
|---|---|---|---|
| combine an autoregressive context model with a hyperprior | Learned image compression · Kodak | rate-distortion performance (PSNR): 8.41% average rate savings over BPG (4:4:4) | Helped |
| combine an autoregressive context model with a hyperprior | Learned image compression · Kodak | rate-distortion performance (MS-SSIM) | Helped |
| add a hyperprior to the autoregressive context-only model (ablation) | Context-only model · Kodak | rate-distortion performance | Helped |
| add an autoregressive context model to the hyperprior (ablation) | Mean & Scale Hyperprior · Kodak | rate-distortion performance | Helped |
| replace the conditional Gaussian distribution with a logistic distribution (ablation) | Context + Hyperprior · Kodak | file size: +0.3% | No clear effect |
| replace the conditional Gaussian distribution with a Laplacian distribution (ablation) | Context + Hyperprior · Kodak | compression performance | Hurt |
| use a 3×3 masked convolution instead of 5×5 (ablation) | Context + Hyperprior · Kodak | rate-distortion performance | Hurt |
| use a 7×7 masked convolution instead of 5×5 (ablation) | Context + Hyperprior · Kodak | rate-distortion performance | Hurt |
| use a 3-layer nonlinear context model instead of a single linear layer (ablation) | Context + Hyperprior · Kodak | rate-distortion performance | Hurt |
| restrict context to a single neighbor (ablation) | Context + Hyperprior · Kodak | rate: +2.1% | Hurt |
| restrict context to three neighbors from the previous row (ablation) | Context + Hyperprior · Kodak | rate: +3.1% | Hurt |
| optimize the model for MSE rather than MS-SSIM (ablation) | Context + Hyperprior · Kodak | MS-SSIM | Hurt |
Closest papers by meaning
- Towards Efficient Image Compression Without Autoregressive Models (NeurIPS 2023 · similarity 0.92)
- Hierarchical Autoregressive Modeling for Neural Video Compression (ICLR 2021 · similarity 0.92)
- Progressive Compression with Universally Quantized Diffusion Models (ICLR 2025 · similarity 0.88)
- Learned Image Compression with Dictionary-based Entropy Model (CVPR 2025 · similarity 0.88)
- Learned Lossless Image Compression based on Bit Plane Slicing (CVPR 2024 · similarity 0.88)
- Hierarchical Masked Autoregressive Models with Low-Resolution Token Pivots (ICML 2025 · similarity 0.88)
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.