Research map / Architectures and efficiency / Attention and transformers
Entroformer: A Transformer-based Entropy Model for Learned Image Compression
Yichen Qian, Xiuyu Sun, Ming Lin, Zhiyu Tan, Rong Jin
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| Replace the CNN entropy model with Entroformer using hyperprior and context | Learned image compression · Kodak, low bit rates | bpp reduction: 5.2% | Improved |
| Replace BPG with Entroformer | Image compression · Kodak, low bit rates | bpp reduction: 20.5% | Improved |
| Use relative 2D diamond position encoding instead of the CNN baseline (ablation) | Entroformer context model · Kodak, λ = 0.02 | bpp: 0.6621 -> 0.6238 | Improved |
| Replace no position encoding with absolute 2D position encoding (ablation) | Transformer context model · Kodak, λ = 0.02 | bpp: 0.6569 -> 0.6563 | Improved |
| Use relative 1D+1D position encoding instead of absolute 2D position encoding (ablation) | Transformer context model · Kodak, λ = 0.02 | bpp: 0.6563 -> 0.6358 | Improved |
| Extend relative position encoding from 1D+1D to 2D (ablation) | Transformer context model · Kodak, λ = 0.02 | bpp: 0.6358 -> 0.6296 | Improved |
| Add a diamond boundary to relative 2D position encoding (ablation) | Transformer context model · Kodak, λ = 0.02 | bpp: 0.6296 -> 0.6238 | Improved |
| Use relative instead of absolute position encoding (ablation) | Entroformer · Tecnick, 300 × 300 | bpp: 0.7654 -> 0.7180 | Improved |
| Use relative instead of absolute position encoding (ablation) | Entroformer · Tecnick, 600 × 600 | bpp: 0.5360 -> 0.5057 | Improved |
| Use relative instead of absolute position encoding (ablation) | Entroformer · Tecnick, 900 × 900 | bpp: 0.4938 -> 0.4154 | Improved |
| Use relative instead of absolute position encoding (ablation) | Entroformer · Tecnick, 1200 × 1200 | bpp: 0.4808 -> 0.4099 | Improved |
| Add bidirectional instead of unidirectional context for the second latent slice (ablation) | Entroformer with hyperprior · Kodak, second latent slice | bpp: 0.3105 -> 0.2972 | Improved |
Closest papers by meaning
- Joint Global and Local Hierarchical Priors for Learned Image Compression (CVPR 2022 · similarity 0.91)
- Conditional Probability Models for Deep Image Compression (CVPR 2018 · similarity 0.89)
- An Information-Theoretic Regularizer for Lossy Neural Image Compression (ICCV 2025 · similarity 0.88)
- Learned Image Compression with Dictionary-based Entropy Model (CVPR 2025 · similarity 0.88)
- Laplacian-guided Entropy Model in Neural Codec with Blur-dissipated Synthesis (CVPR 2024 · similarity 0.87)
- Robustly overfitting latents for flexible neural image compression (NeurIPS 2024 · 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.
