Research map / Image generation / Image restoration
Task-Aware Image Downscaling
Heewon Kim, Myungsub Choi, Bee Lim, Kyoung Mu Lee
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 |
|---|---|---|---|
| retrain EDSR+ on learned TAD-HR pairs instead of bicubic LR-HR pairs | EDSR+ · Set5 x2 (color channel) | PSNR (dB): 36.09 -> 39.44 | Improved |
| retrain EDSR+ on TAD-HR pairs | EDSR+ · Urban100 x4 (color channel) | PSNR (dB): 25.34 -> 27.76 | Improved |
| retrain EDSR+ on TAD-HR pairs | EDSR+ · DIV2K val x2 (color channel) | PSNR (dB): 35.12 -> 40.21 | Improved |
| retrain EDSR+ on TAD-HR pairs | EDSR+ · Set5 x2 (single channel) | PSNR (dB): 37.95 -> 37.98 | No clear effect |
| retrain EDSR+ on TAD-HR pairs | EDSR+ · Urban100 x2 (single channel) | PSNR (dB): 32.51 -> 32.86 | Improved |
| jointly train downscaler and upscaler (TAD-TAU) instead of bicubic downscaling with the same upscaler | TAU · Urban100 x2 (color channel) | PSNR (dB): 29.13 -> 35.03 | Improved |
| use TAD-TAU rather than bicubic-trained EDSR+ at smaller network depth | EDSR+ · B100 x2 (color channel) | PSNR (dB): 31.40 -> 36.68 | Improved |
| learn grayscale conversion jointly with colorizer instead of standard grayscale | colorization network · B100 | PSNR (dB): 24.21 -> 36.14 | Improved |
| learn grayscale conversion jointly with colorizer | colorization network · Urban100 | PSNR (dB): 23.29 -> 33.68 | Improved |
| use TAU instead of EDSR+ for upscaling a full HD image (runtime) | EDSR+ · Full HD upscaling | seconds: 70.88 -> 0.14 | Improved |
| increase guidance weight lambda from 0 toward 100 (ablation) | TAD-TAU · DIV2K val (10 images) x2 | PSNR (dB) | Mixed |
Closest papers by meaning
- Downscaled Representation Matters: Improving Image Rescaling with Collaborative Downscaled Images (ICCV 2023 · similarity 0.89)
- Towards Bidirectional Arbitrary Image Rescaling: Joint Optimization and Cycle Idempotence (CVPR 2022 · similarity 0.89)
- Deep Generative Model based Rate-Distortion for Image Downscaling Assessment (CVPR 2024 · similarity 0.87)
- Block-Based Multi-Scale Image Rescaling (AAAI 2025 · similarity 0.86)
- Debiased All-in-one Image Restoration with Task Uncertainty Regularization (AAAI 2025 · similarity 0.86)
- Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss (CVPR 2024 · similarity 0.85)
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.
