Research map / Image generation / Super-resolution
Hierarchical Image Tokenization for Multi-Scale Image Super Resolution
Isma Hadji, Enrique Sanchez, Adrian Bulat, Brais Martinez, Georgios Tzimiropoulos
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 |
|---|---|---|---|
| add DPO regularization (ablation) | H-VAR · RealSR (128 output scale) | PSNR: 20.56 -> 22.09 | Improved |
| add DPO regularization (ablation) | H-VAR · RealSR (128 output scale) | LPIPS: 0.347 -> 0.1996 | Improved |
| add DPO regularization (ablation) | H-VAR · DRealSR (128 output scale) | PSNR: 23.03 -> 25.26 | Improved |
| add DPO regularization (ablation) | H-VAR · DRealSR (128 output scale) | LPIPS: 0.311 -> 0.1871 | Improved |
| add DPO regularization (ablation) | H-VAR · RealSR (256 output scale) | PSNR: 23.09 -> 24.41 | Improved |
| add DPO regularization (ablation) | H-VAR · RealSR (512 output scale) | PSNR: 25.72 -> 25.55 | No clear effect |
| add DPO regularization (ablation) | H-VAR · DRealSR (512 output scale) | PSNR: 28.61 -> 28.73 | No clear effect |
| add hierarchical image tokenization (H-VAR vs VARSR-d16 same backbone and data) (ablation) | VARSR-d16 · DIV2K-Val | LPIPS: 0.495 -> 0.317 | Improved |
| add hierarchical image tokenization (H-VAR vs VARSR-d16 same backbone and data) (ablation) | VARSR-d16 · RealSR | FID: 231.1 -> 130.4 | Improved |
| add hierarchical image tokenization (H-VAR vs VARSR-d16 same backbone and data) (ablation) | VARSR-d16 · DRealSR | PSNR: 27.30 -> 28.73 | Improved |
| use 310M H-VAR instead of 1B VARSR trained on large proprietary data | VARSR · RealSR | LPIPS: 0.350 -> 0.256 | Improved |
| use 310M H-VAR instead of 1B VARSR | VARSR · DIV2K-Val | MUSIQ: 71.27 -> 69.40 | Hurt |
Closest papers by meaning
- Visual Autoregressive Modeling for Image Super-Resolution (ICML 2025 · similarity 0.91)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.88)
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · similarity 0.88)
- Channel-wise Vector Quantization (NeurIPS 2026 · similarity 0.88)
- An Image is Worth 32 Tokens for Reconstruction and Generation (NeurIPS 2024 · similarity 0.88)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (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.
