Research map / Image generation / Image restoration
Learned image compression
Compress images with learned hyperprior entropy models and lossless or lossy coding.
69 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Image restoration, keyword cluster “compression · distortion · coding”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
10 papers in 2023–24 and 29 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.2% to 6.7%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · End-to-end Optimized Image Compression (ICLR 2017 · 1,342 citations)
- 2018 · Variational image compression with a scale hyperprior (ICLR 2018 · 1,378 citations)
- 2019 · Practical Full Resolution Learned Lossless Image Compression (CVPR 2019 · 228 citations)
- 2020 · High-Fidelity Generative Image Compression (NeurIPS 2020 · 211 citations)
- 2021 · Deep Homography for Efficient Stereo Image Compression (CVPR 2021 · 55 citations)
- 2022 · Neural Data-Dependent Transform for Learned Image Compression (CVPR 2022 · 59 citations)
- 2023 · Lossy Image Compression with Conditional Diffusion Models (NeurIPS 2023 · 30 citations)
- 2024 · Make Lossy Compression Meaningful for Low-Light Images (AAAI 2024 · 9 citations)
- 2025 · Learned Image Compression with Dictionary-based Entropy Model (CVPR 2025 · 23 citations)
- 2026 · MRT: Learning Compact Representations with Mixed RWKV-Transformer for Extreme Image Compression (AAAI 2026 · 1 citation)
Earliest papers here
- End-to-end Optimized Image Compression (ICLR 2017 · 1,342 citations)
- Lossy Image Compression with Compressive Autoencoders (ICLR 2017 · 256 citations)
- Real-Time Adaptive Image Compression (ICML 2017 · 111 citations)
What has been tried
49 papers here report 458 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 5%, hurt 27%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add the dictionary-based cross-attention entropy module (DCAE) | Kodak · BD-rate | -4.20% -> -7.28% | Helped | Learned Image Compression with Dictionary-based Entropy Model (CVPR 2025 · 23 citations) |
| add spatial grouping | ImageNet, λ=64 · CBR | 0.0590 -> 0.0403 | Helped | Learned Image Transmission with Hierarchical Variational Autoencoder (AAAI 2025 · 10 citations) |
| add an SNR-aware branch to extract local and non-local information | SID · rate-distortion performance | Helped | Make Lossy Compression Meaningful for Low-Light Images (AAAI 2024 · 9 citations) | |
| Replace standard 4×4 window attention and conventional FFN with a larger 16×16 window | Kodak · BD-rate vs VTM-12.1 | -9.3% -> -11.2% | Helped | Frequency-Aware Transformer for Learned Image Compression (ICLR 2024 · 7 citations) |
| Use saliency-derived σ instead of constant σ | CLIC2020 professional validation, high bit rates · visual quality | Helped | Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion (CVPR 2025 · 5 citations) | |
| Convert multi-channel images into a one-channel 2D matrix before entropy modeling (one-channel scheme) | Kodak · RD performance (PSNR vs bpp) | Helped | GLIC: General Format Learned Image Compression (AAAI 2025 · 4 citations) | |
| Increase the number of reference images from 1 to 3 | Kodak · BD-Rate_P | -10.2 -> -14.5 | Helped | Conditional Latent Coding with Learnable Synthesized Reference for Deep Image Compression (AAAI 2025 · 3 citations) |
| Use diffusion compensation during decoding instead of removing the diffusion module | fMoW test set · LPIPS | Helped | COSMIC: Compress Satellite Image Efficiently via Diffusion Compensation (NeurIPS 2024 · 3 citations) | |
| replace CAMSIC entropy model with ECSIC entropy model | Cityscapes · BD-rate | 0% -> 48.89% | Hurt | CAMSIC: Content-aware Masked Image Modeling Transformer for Stereo Image Compression (AAAI 2025 · 2 citations) |
| Add overfitting to the latent variable model | Kodak · BPD | 3.36 -> 2.88 | Helped | Fitted Neural Lossless Image Compression (CVPR 2025 · 2 citations) |
| add hierarchical linear projection to the LZMA baseline | C16 patches · compression ratio | 1.25 -> 2.12 | Helped | WISE: A Framework for Gigapixel Whole-Slide-Image Lossless Compression (CVPR 2025 · 1 citation) |
| Remove the orthogonality constraint for OLP | Kodak · rate-distortion performance | Hurt | On Disentangled Training for Nonlinear Transform in Learned Image Compression (ICLR 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 15 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Flickr2W · 2 of 15 papers
- Kodak · 2 of 15 papers
- Open Images · 2 of 15 papers
Baselines
- ELIC · 4 of 15 papers
- BPG · 3 of 15 papers
- MLIC+ · 3 of 15 papers
- JPEG2000 · 3 of 15 papers
- VVC · 2 of 15 papers
- VTM · 2 of 15 papers
- JPEG · 2 of 15 papers
- AV1 · 2 of 15 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 39 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- Korea University · 2 papers
- Samsung · 2 papers
- Nanyang Technological University · 2 papers
- Peng Cheng Laboratory · 2 papers
- Harbin Institute of Technology · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Perceptual diffusion image compression (28 papers)
Elsewhere in the topic
- Dehazing and deraining (118 papers)
- Efficient restoration architectures (113 papers)
- Inverse imaging with learned priors (109 papers)
- Image harmonization and color transformation (102 papers)
- Reflection removal (95 papers)
- Pansharpening and spectral fusion (84 papers)
- Image deblurring (73 papers)
- Low-light image enhancement (72 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).