Research map / Image generation / Image restoration
Self-Supervised Burst Super-Resolution
Goutam Bhat, Michaël Gharbi, Jiawen Chen, Luc Van Gool, Zhihao Xia
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 |
|---|---|---|---|
| train with the self-supervised burst objective on real bursts instead of weakly-paired supervision | Burstormer · SynBSR (4x) | PSNR (dB): 36.84 -> 38.81 | Improved |
| train with self-supervised objective instead of synthetic supervised data | DeepRep · SynBSR (4x) | PSNR (dB): 36.53 -> 38.73 | Improved |
| weakly-paired supervision vs synthetic supervision | DeepRep · SynBSR (4x) | PSNR (dB): 36.53 -> 34.62 | Hurt |
| learn the blur kernel jointly instead of fixing an incorrect kernel (ablation) | self-supervised training · Synthetic 2x, unit impulse blur | PSNR (dB): 28.328 -> 38.451 | Improved |
| learn the blur kernel jointly on anisotropic Gaussian blur (ablation) | self-supervised training · Synthetic 2x, anisotropic Gaussian blur | PSNR (dB): 34.822 -> 36.205 | Mixed |
| estimate motion with optical flow instead of using known motion (ablation) | self-supervised training · Synthetic 2x, unit impulse blur | PSNR (dB): 40.359 -> 38.999 | Hurt |
| add validity mask to the reconstruction loss (ablation) | self-supervised training · BurstSR (qualitative) | ghosting artifacts | Improved |
| self-supervised training versus fully supervised burst SR with known blur and motion | DeepRep (small) · Synthetic 2x, five data distributions | PSNR (dB): worst case 0.55 dB lower than supervised | Mixed |
| fine-tune on HDR+ bursts without ground truth | DeepRep · HDR+ Nexus 6 (qualitative) | visual quality | Improved |
Closest papers by meaning
- Deep Burst Super-Resolution (CVPR 2021 · similarity 0.90)
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · similarity 0.88)
- Burst Image Restoration and Enhancement (CVPR 2022 · similarity 0.88)
- Towards Real-World Burst Image Super-Resolution: Benchmark and Method (ICCV 2023 · similarity 0.88)
- QMambaBSR: Burst Image Super-Resolution with Query State Space Model (CVPR 2025 · similarity 0.87)
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · 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.
