Research map / Detection and segmentation / Open-set and few-shot detection
Learning Transformations To Reduce the Geometric Shift in Object Detection
Vidit Vidit, Martin Engilberge, Mathieu Salzmann
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 |
|---|---|---|---|
| Learn homographies and aggregate their transformed features with target-domain self-training instead of using MT alone | FasterRCNN ResNet50 · Cityscapes → KITTI FoV adaptation | Car AP@0.5: 78.3 -> 80.4 | Improved |
| Use learned transformations instead of PIT correction | Ours vs. MT+PIT · Cityscapes → KITTI FoV adaptation | Car AP@0.5: 79.7 -> 80.4 | Improved |
| Use Mean Teacher pseudo-label adaptation rather than source-only training | FasterRCNN ResNet50 · Cityscapes → KITTI FoV adaptation | Car AP@0.5: 76.1 -> 78.3 | Improved |
| Optimize homographies instead of using FR+PIT (ablation) | Ours-h · FoVx 70° | Car AP@0.5: 91.4 -> 93.1 | Improved |
| Optimize homographies instead of using FR+PIT (ablation) | Ours-h · FoVx 80° | Car AP@0.5: 89.2 -> 91.8 | Improved |
| Optimize homographies instead of using FR+PIT (ablation) | Ours-h · FoVx 90° | Car AP@0.5: 85.9 -> 88.8 | Improved |
| Optimize homographies instead of using FR (ablation) | Ours-h · FoVx 50° | Car AP@0.5: 94.3 -> 94.1 | Hurt |
| Optimize only homography parameters, keeping the base and aggregator frozen (ablation) | FasterRCNN ResNet50 · Cityscapes → KITTI FoV adaptation | AP@0.5: FR 76.1; Ours-h 78.2 | Improved |
| Optimize only homography parameters, keeping the base and aggregator frozen (ablation) | FasterRCNN ResNet50 · Cityscapes → MOT viewpoint adaptation | AP@0.5: 49.8; FR value not reported in the supplied text | Improved |
| Initialize all homographies to identity and learn a diverse set (ablation) | Ours · Cityscapes → KITTI FoV adaptation | AP@0.5: 79.5 AP@0.5 | Improved |
| Increase the number of learned homographies (ablation) | Ours · Cityscapes → KITTI FoV adaptation | AP | Improved |
Closest papers by meaning
- Analysing Domain Shift Factors between Videos and Images for Object Detection (TPAMI 2016 · similarity 0.86)
- GOOD: Exploring geometric cues for detecting objects in an open world (ICLR 2023 · similarity 0.86)
- GOLDILOCS: GENERAL OBJECT-LEVEL DETECTION AND LABELING OF CHANGES IN SCENES (ICLR 2026 · similarity 0.86)
- Objects Can Move: 3D Change Detection by Geometric Transformation Consistency (ECCV 2022 · similarity 0.86)
- Learned Two-Plane Perspective Prior Based Image Resampling for Efficient Object Detection (CVPR 2023 · similarity 0.85)
- Multi-View Pose-Agnostic Change Localization with Zero Labels (CVPR 2025 · 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.
