Research map / Image generation / Image restoration
Learning to Inpaint for Image Compression
Mohammad Haris Baig, Vladlen Koltun, Lorenzo Torresani
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 |
|---|---|---|---|
| Predict the original image at each stage instead of predicting only residuals | Progressive residual encoder on Kodak · Kodak | file size at similar quality: 18% | Improved |
| Add inpainting to the R2I compression model | R2I on Kodak · Kodak | file size at similar quality: 42% | Improved |
| Jointly train inpainting with the R2I compression model instead of training them separately (ablation) | R2I decoding connections · Kodak | rate savings | Improved |
| Use decoding connections rather than full connections between stages (ablation) | R2I · Kodak | rate savings | Improved |
| Use decoding connections rather than sharing only the final prediction (ablation) | R2I · Kodak | rate savings | Improved |
| Add multi-scale convolutions to the inpainting network (ablation) | Partial-context inpainting on Kodak · Kodak partial-context inpainting | SSIM: 0.4545 -> 0.5165 | Improved |
| Replace learned multi-scale inpainting with a vanilla CNN without multi-scale convolutions (ablation) | Partial-context inpainting on Kodak · Kodak partial-context inpainting | SSIM: 0.5165 -> 0.4545 | Hurt |
| Use inpainting instead of no inpainting (ablation) | R2I decoding connections on Kodak · Kodak | MS-SSIM (dB): 11.19 -> 10.37; 11.19 -> 11.52 | Mixed |
Closest papers by meaning
- How To Exploit the Transferability of Learned Image Compression to Conventional Codecs (CVPR 2021 · similarity 0.88)
- Real-Time Adaptive Image Compression (ICML 2017 · similarity 0.88)
- Learning End-to-End Lossy Image Compression: A Benchmark (TPAMI 2021 · similarity 0.88)
- End-to-end Optimized Image Compression (ICLR 2017 · similarity 0.87)
- Lossy Image Compression with Compressive Autoencoders (ICLR 2017 · similarity 0.87)
- GLIC: General Format Learned Image Compression (AAAI 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.
