Research map / Detection and segmentation / Weakly supervised detection and segmentation
Active Learning Strategies for Weakly-Supervised Object Detection
Huy V. Vo, Oriane Siméoni, Spyros Gidaris, Andrei Bursuc, Patrick Pérez, Jean Ponce
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 |
|---|---|---|---|
| replace uniform random selection with BiB selection | MIST + few full annotations · VOC07 (250 images) | AP50: 61.1 -> 63.5 | Improved |
| add difficulty-aware proposal sampling (DifS) to random selection (ablation) | u-random · VOC07 (250 images) | AP50: 60.5 -> 61.1 | Improved |
| add k-means++ selection on image-level features (ablation) | DifS + random · VOC07 (250 images) | AP50: 61.1 -> 60.3 | Hurt |
| add k-means++ selection on region-level features (ablation) | DifS + random · VOC07 (250 images) | AP50: 61.1 -> 63.4 | Improved |
| use BiB pairs instead of generic region features for diverse selection (ablation) | DifS + K selection · VOC07 (250 images) | AP50: 63.4 -> 63.5 | Improved |
| add DifS to BiB (ablation) | BiB · VOC07 (250 images) | AP50: 62.6 -> 63.5 | Improved |
| add DifS to BiB (ablation) | BiB · COCO (800 images) | AP50: 34.2 -> 34.1 | No clear effect |
| add diversity sampling to BiB-pair image selection (ablation) | BiB · VOC07 (cycle 5) | AP50: 61.8 -> 63.5 | Improved |
| use BiB selection instead of loss-based selection | MIST + few full annotations · COCO (800 images) | AP50: 30.4 -> 34.1 | Improved |
| use entropy-sum selection instead of uniform random | MIST + few full annotations · COCO (800 images) | AP50: 33.0 -> 27.8 | Hurt |
| balance classes in random selection | u-random · VOC07 (250 images) | AP50: 61.1 -> 61.2 | No clear effect |
| vary area ratio mu from 2 to 4 (ablation) | BiB · VOC07 (250 images) | AP50: 63.1 -> 63.5 -> 63.3 | No clear effect |
Closest papers by meaning
- Box-Level Active Detection (CVPR 2023 · similarity 0.91)
- Track and Transfer: Watching Videos to Simulate Strong Human Supervision for Weakly-Supervised Object Detection (CVPR 2016 · similarity 0.91)
- W2F: A Weakly-Supervised to Fully-Supervised Framework for Object Detection (CVPR 2018 · similarity 0.90)
- DLWL: Improving Detection for Lowshot Classes With Weakly Labelled Data (CVPR 2020 · similarity 0.90)
- ALWOD: Active Learning for Weakly-Supervised Object Detection (ICCV 2023 · similarity 0.90)
- Many-shot from Low-shot: Learning to Annotate using Mixed Supervision for Object Detection (ECCV 2020 · similarity 0.90)
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.
