Research map / Detection and segmentation / Open-set and few-shot detection
Möbius Transform for Mitigating Perspective Distortions in Representation Learning
Prakash Chandra Chhipa, Meenakshi Subhash Chippa, Kanjar De, Rajkumar Saini, Marcus Liwicki, Mubarak Shah
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 MPD augmentation (P=0.6) to supervised ResNet50 training (ablation) | ResNet50 · ImageNet-PD PD-B | Top-1 accuracy: 61.15 -> 73.01 | Improved |
| add MPD augmentation (P=0.6) to supervised ResNet50 training (ablation) | ResNet50 · ImageNet val (original) | Top-1 accuracy: 76.13 -> 76.19 | No clear effect |
| apply MPD with probability 1.0 instead of 0.8 (ablation) | supervised:MPD · ImageNet val (original) | Top-1 accuracy: 76.34 -> 74.10 | Hurt |
| apply MPD IB with probability 1.0 instead of 0.8 (ablation) | supervised:MPD IB · ImageNet val (original) | Top-1 accuracy: 75.63 -> 74.62 | Hurt |
| train ResNet50 with MPD augmentation | ResNet50 · ImageNet-X | accuracy: 60.64 -> 63.23 | Improved |
| train ResNet50 with MPD augmentation | ResNet50 · ImageNet-E average | Top-1 accuracy drop: 15.72 -> 13.34 | Improved |
| train ResNet50 with MPD augmentation on background-change subsets | ResNet50 · ImageNet-E background changes | Top-1 accuracy | No clear effect |
| add MPD-AutoCrowd augmentation to P2P-Net | P2P-Net · UCF-CC-50 | MAE: 172.72 -> 96.80 | Improved |
| add MPD-AutoCrowd augmentation to P2P-Net | P2P-Net · ShanghaiTech Part B | MAE: 6.25 -> 6.61 | Mixed |
| add MPD to Clip-ReIdent | Clip-ReIdent ResNet50x16 · DeepSportRadar re-id | mAP (w/o RR): 88.50 -> 91.95 | Improved |
| add MPD to Clip-ReIdent | Clip-ReIdent ViT-L-14 · DeepSportRadar re-id | mAP (w/o RR): 96.90 -> 97.02 | Improved |
| add MPD-OD augmentation to FasterRCNN and FCOS at P=0.5 | FasterRCNN, FCOS · COCO | detection accuracy: 3% and 1.6% gain | Improved |
Closest papers by meaning
- Learning Perspective Undistortion of Portraits (ICCV 2019 · similarity 0.84)
- MPED: Quantifying Point Cloud Distortion Based on Multiscale Potential Energy Discrepancy (TPAMI 2022 · similarity 0.84)
- From Perspective to Fisheye Depth Estimation and Open-Vocabulary Segmentation (ECCV 2026 · similarity 0.84)
- DarSwin: Distortion Aware Radial Swin Transformer (ICCV 2023 · similarity 0.84)
- Discriminating image representations with principal distortions (ICLR 2025 · similarity 0.84)
- PanDA: Towards Panoramic Depth Anything with Unlabeled Panoramas and Mobius Spatial Augmentation (CVPR 2025 · similarity 0.82)
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.
