Research map / Image generation / Image restoration
Self-Asymmetric Invertible Network for Compression-Aware Image Rescaling
Jinhai Yang, Mengxi Guo, Shijie Zhao, Junlin Li, Li Zhang
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 |
|---|---|---|---|
| Use compression-aware asymmetric invertible rescaling | Image rescaling · DIV2K validation, JPEG QF=30 and 90, ×2 and ×4 | PSNR: 31.47 vs 29.38; 35.96 vs 33.44; 27.90 vs 26.27; 30.31 vs 28.98 | Improved |
| Use compression-aware asymmetric invertible rescaling | Image rescaling · DIV2K, WebP QF=30 and 90, ×2 | PSNR: 33.15 vs 29.86; 35.83 vs 34.34 | Improved |
| Jointly optimize downscaling and upscaling using the asymmetric framework rather than compression-aware fine-tuning (ablation) | IRN & EDSR · DIV2K, JPEG QF=75, ×2 | PSNR: 32.92 -> 34.41 | Improved |
| Replace the Dual-IRN framework with SAIN (ablation) | Compression-aware asymmetric image rescaling · DIV2K, JPEG QF=75, ×2 | PSNR: 34.69 -> 35.10 | Improved |
| Add a learnable Gaussian mixture instead of using N(0, 1) (ablation) | SAIN · DIV2K, JPEG QF=75, ×2 | PSNR: 35.01 -> 35.10 | Improved |
| Add a learnable Gaussian mixture instead of using N(0, 1) (ablation) | SAIN · Set5, JPEG QF=75, ×2 | PSNR: 35.98 -> 36.04 | Improved |
| Add the L_rel loss (ablation) | SAIN · DIV2K, JPEG QF=75, ×2 | PSNR: 35.02 -> 35.10 | Improved |
| Add the L_fit' loss to guide compression simulation (ablation) | SAIN · DIV2K, JPEG QF=75, ×2 | PSNR: 34.95 -> 35.10 | Improved |
| Add the L_reg loss (ablation) | SAIN · DIV2K, JPEG QF=75, ×2 | PSNR: 34.96 -> 35.10 | Improved |
| Use five E-InvBlocks in the downscaling module (ablation) | SAIN · DIV2K, JPEG QF=75, ×2 | PSNR | Improved |
| Use five Gaussian mixture components (ablation) | SAIN · DIV2K, JPEG QF=75, ×2 | PSNR | Improved |
| Train at QF=75 rather than QF=25 (ablation) | SAIN · DIV2K, JPEG QF=25 evaluation, ×2 | PSNR: 29.14 -> 30.89 | Hurt |
Closest papers by meaning
- Invertible Rescaling Network and Its Extensions (IJCV 2022 · similarity 0.91)
- Invertible Image Rescaling (ECCV 2020 · similarity 0.90)
- Plug-and-Play Tri-Branch Invertible Block for Image Rescaling (AAAI 2025 · similarity 0.90)
- Deep Generative Model based Rate-Distortion for Image Downscaling Assessment (CVPR 2024 · similarity 0.86)
- Ultra Lowrate Image Compression with Semantic Residual Coding and Compression-aware Diffusion (ICML 2025 · similarity 0.86)
- Block-Based Multi-Scale Image Rescaling (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.
