Research map / Image generation / Super-resolution
Accelerating Image Super-Resolution Networks with Pixel-Level Classification
Jinho Jeong, Jinwoo Kim, Younghyun Jo, Seon Joo Kim
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 patch-level allocation with pixel-level classification | CARN · Test2K x4 | GFLOPs: 99.8 (ARM) -> 64.0 at similar PSNR | Improved |
| increase patch size for patch-level ClassSR versus PCSR (ablation) | CARN-ClassSR · Test2K x4 | GFLOPs: 66.7 -> 75.8 | Hurt |
| increase patch size for PCSR (ablation) | CARN-PCSR · Test2K x4 | GFLOPs: 61.1 -> 54.5 | Improved |
| use 3 classes instead of 2 (ablation) | CARN-PCSR · Test2K x4 | GFLOPs: 64.0 -> 62.4 | No clear effect |
| add pixel-wise refinement (#h=1) to reduce artifacts (ablation) | CARN-PCSR · Test2K x4 | PSNR: 26.022 -> 25.995 (#h=9 vs #h=0) | No clear effect |
| use LIIF upsampler instead of pixel-shuffle (ablation) | SRResNet · Urban100 x4 | PSNR: 24.65 -> 24.63 | No clear effect |
| apply PCSR to BSRN lightweight model | BSRN · Urban100 x4 | PSNR: 24.43 -> 24.29, GFLOPs 29.1 -> 23.6 | Mixed |
| compare to per-image method MGA | FSRCNN-PCSR · Test4K x4 | GFLOPs: 101.7 -> 37.8 | Improved |
| apply PCSR on small images (Urban100) with ClassSR/ARM | FSRCNN-ClassSR · Urban100 x4 | GFLOPs: 19.9 -> 20.8 (105%) | Hurt |
Closest papers by meaning
- ClassSR: A General Framework to Accelerate Super-Resolution Networks by Data Characteristic (CVPR 2021 · similarity 0.90)
- Fast Adaptation to Super-Resolution Networks via Meta-Learning (ECCV 2020 · similarity 0.90)
- Adaptive Patch Exiting for Scalable Single Image Super-Resolution (ECCV 2022 · similarity 0.89)
- Pixel Recursive Super Resolution (ICCV 2017 · similarity 0.89)
- Accelerating the Super-Resolution Convolutional Neural Network (ECCV 2016 · similarity 0.89)
- SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network (AAAI 2024 · similarity 0.88)
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.
