Research map / Image generation / Super-resolution
Exploring Resolution and Degradation Clues As Self-Supervised Signal for Low Quality Object Detection
Ziteng Cui, Yingying Zhu, Lin Gu, Guo-Jun Qi, Xiaoxiao Li, Renrui Zhang, Zenghui Zhang, Tatsuya Harada
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 degradation-transformation training (Deg t) (ablation) | CenterNet ResNet-18 · COCO-d | mAP: 14.5 -> 16.8 (+2.3) | Improved |
| add feature connection (skip links at /8 and /16) (ablation) | CenterNet + Deg t · COCO-d | mAP: 16.8 -> 17.5 | Improved |
| add ARRD at middle up-sampling layer (loc 2) (ablation) | CenterNet + Deg t + feature connect · COCO-d | mAP: 17.5 -> 18.4 | Improved |
| place ARRD at shallow layer loc 1 instead of loc 2 (ablation) | AERIS · COCO-d | mAP: 18.4 -> 18.2 | Hurt |
| place ARRD at final layer loc 3 instead of loc 2 (ablation) | AERIS · COCO-d | mAP: 18.4 -> 17.9 | Hurt |
| AERIS versus BSRGAN x2 pre-processing with Swin-T backbone | CenterNet Swin-T · COCO-d | AP: 20.2 -> 21.6 | Improved |
| pre-process with Restormer denoising on multi-degradation images | CenterNet ResNet-18 · COCO-d | AP: 14.5 -> 11.4 | Hurt |
| add ARRD to RetinaNet-FPN Swin-T on all FPN levels (1,2,3,4) (ablation) | RetinaNet-FPN Swin-T · COCO-d | AP: 21.5 -> 22.1 | Improved |
| add ARRD on FPN level 1 only (ablation) | RetinaNet-FPN Swin-T · COCO-d | AP: 21.5 -> 22.0 | Improved |
| fine-tune BSRGAN on multi-degradation data | BSRGAN pre-process · COCO-d | AP: 16.8 -> 16.9 | Improved |
| fine-tune DBPN on multi-degradation data | DBPN pre-process · COCO-d | AP: 15.0 -> 16.4 | Improved |
| train with down-sampling ratio range (1,2) instead of (1,4) (ablation) | AERIS · COCO-d | AP: 18.4 -> 17.8 | Hurt |
Closest papers by meaning
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · similarity 0.87)
- Unsupervised Degradation Representation Aware Transform for Real-World Blind Image Super-Resolution (AAAI 2025 · similarity 0.85)
- Learning Controllable Degradation for Real-World Super-Resolution via Constrained Flows (ICML 2023 · similarity 0.85)
- Better to Follow, Follow to Be Better: Towards Precise Supervision of Feature Super-Resolution for Small Object Detection (ICCV 2019 · similarity 0.85)
- LightBSR: Towards Lightweight Blind Super-Resolution via Discriminative Implicit Degradation Representation Learning (ICCV 2025 · similarity 0.85)
- Unsupervised Degradation Representation Learning for Blind Super-Resolution (CVPR 2021 · 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.
