Research map / Detection and segmentation / Weakly supervised detection and segmentation
ALWOD: Active Learning for Weakly-Supervised Object Detection
Yuting Wang, Velibor Ilic, Jiatong Li, Branislav Kisačanin, Vladimir Pavlovic
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 auxiliary generated fully-labeled set S^X to the initial student-teacher training (ablation) | ALWOD (VGG16) · RealPizza10 initial detector | AP50: 11.9 -> 18.8 | Improved |
| add auxiliary generated fully-labeled set S^X to the initial student-teacher training (ablation) | ALWOD (VGG16) · VOC2007 initial detector | AP50: 55.4 -> 60.1 | Improved |
| add auxiliary set S^X when fine-tuning the MIST base detector in BiB (ablation) | BiB (VGG16, MIST) · VOC2007 and RealPizza10 initial detector | AP50: 54.5 -> 50.6 (VOC2007); 15.2 -> 14.0 (RealPizza10) | Hurt |
| use product of model disagreement and image uncertainty instead of sum (ablation) | ALWOD acquisition function · VOC2007, COCO2014, RealPizza10 | AP50 | Improved |
| use ALWOD acquisition function instead of uniform sampling (ablation) | ALWOD (ResNet50) · VOC2007, COCO2014, RealPizza10 | AP50 | Improved |
| use core-set or loss-prediction acquisition instead of uniform sampling (ablation) | ALWOD framework · VOC2007, COCO2014, RealPizza10 | AP50 | Hurt |
| use entropy-sum acquisition instead of the proposed functions (ablation) | ALWOD framework · RealPizza10, COCO2014 | AP50 | Hurt |
| use selecting-the-box annotation instead of drawing the box (ablation) | ALWOD (ResNet50) · RealPizza10 | AP50: 39.9 -> 39.3 | No clear effect |
| use selecting-the-box annotation instead of drawing the box | ALWOD (ResNet50) · annotation time per image | annotation time: reduction of over 70%, about 2 seconds | Improved |
| replace Faster-RCNN with Sparse DETR as student and teacher detector (ablation) | ALWOD · VOC2007, COCO2014, RealPizza10 | AP50: 69.1 -> 70.5 (VOC2007, ResNet50) | No clear effect |
| use ALWOD with 5% fully annotated data instead of SSDGMM with 80% fully annotated data | RealPizza10 VGG16 · RealPizza10 | AP50: 23.4 -> 37.9 | Improved |
| use a stronger backbone (VGG16 to Swin-T) in ALWOD (ablation) | ALWOD Sparse DETR · VOC2007 | AP50: 68.0 -> 71.7 | Improved |
Closest papers by meaning
- UWSOD: Toward Fully-Supervised-Level Capacity Weakly Supervised Object Detection (NeurIPS 2020 · similarity 0.91)
- W2N: Switching from Weak Supervision to Noisy Supervision for Object Detection (ECCV 2022 · similarity 0.90)
- Active Learning Strategies for Weakly-Supervised Object Detection (ECCV 2022 · similarity 0.90)
- Many-shot from Low-shot: Learning to Annotate using Mixed Supervision for Object Detection (ECCV 2020 · similarity 0.90)
- Portable Active Learning for Object Detection (CVPR 2026 · similarity 0.90)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.89)
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.
