Research map / Image generation / Super-resolution
HiT-SR: Hierarchical Transformer for Efficient Image Super-Resolution
XIANG ZHANG, Yulun Zhang, Fisher Yu
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 |
|---|---|---|---|
| convert SwinIR-Light to hierarchical transformer with SCC (HiT-SIR) | SwinIR-Light · Urban100 x2 | PSNR: 32.76 -> 33.02 | Improved |
| replace shifted window self-attention with SCC | SwinIR-Light · A100 runtime, x2, 720x1280 output | runtime (ms): 1899 -> 249 | Improved |
| convert SRFormer-Light to HiT-SRF | SRFormer-Light · A100 runtime, x2, 720x1280 output | runtime (ms): 2593 -> 331 | Improved |
| use expanding hierarchical windows instead of fixed small windows (ablation) | HiT-SIR · Manga109 x2 | PSNR: 38.88 -> 39.21 | Improved |
| use hierarchical windows instead of fixed large 64x64 windows (ablation) | HiT-SIR · Manga109 x2 | PSNR: 39.09 -> 39.21; FLOPs 241.4G -> 209.9G | Improved |
| expand windows small to large rather than shrink large to small (ablation) | HiT-SIR · Manga109 x2 | PSNR: 39.12 -> 39.21 | Improved |
| use dual feature extraction (linear + conv branches) instead of linear projection (ablation) | HiT-SIR · Manga109 x2 | PSNR: 38.99 -> 39.21 | Improved |
| use queries and values only, estimating no separate keys (ablation) | HiT-SIR · Manga109 x2 | PSNR: 39.19 -> 39.21; params 826K -> 772K | No clear effect |
| use correlation instead of softmax attention (ablation) | HiT-SIR · Manga109 x2 | PSNR: 39.08 -> 39.21 | Improved |
| combine spatial and channel self-correlation rather than either alone (ablation) | HiT-SIR · Manga109 x2 | PSNR: 39.04 (S-SC only), 39.18 (C-SC only) -> 39.21 | Improved |
| multi-head for S-SC and single-head for C-SC instead of uniform head strategy (ablation) | HiT-SIR · Manga109 x2 | PSNR: 39.14 (all single) and 39.08 (all multi) -> 39.21 | Improved |
| HiT-SIR with hierarchical windows trains faster | SwinIR-Light · Urban100 x2, Manga109 x2 | iterations to match performance | Improved |
Closest papers by meaning
- Transcending the Limit of Local Window: Advanced Super-Resolution Transformer with Adaptive Token Dictionary (CVPR 2024 · similarity 0.90)
- Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · similarity 0.90)
- CFAT: Unleashing Triangular Windows for Image Super-resolution (CVPR 2024 · similarity 0.88)
- Effective Diffusion Transformer Architecture for Image Super-Resolution (AAAI 2025 · similarity 0.88)
- Transformer-Based Selective Super-resolution for Efficient Image Refinement (AAAI 2024 · similarity 0.88)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (AAAI 2024 · similarity 0.87)
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.
