Research map / Detection and segmentation / Weakly supervised detection and segmentation
Cyclic-Bootstrap Labeling for Weakly Supervised Object Detection
Yufei Yin, Jiajun Deng, Wengang Zhou, Li Li, Houqiang Li
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 WET teacher and CRD rank distillation to the basic WSOD module (ablation) | Basic WSOD (VGG16) · PASCAL VOC 2007 test | mAP: 53.3 -> 55.8 | Improved |
| use WET as distillation teacher instead of the last OIC branch (ablation) | CRD · PASCAL VOC 2007 test | mAP: 54.2 -> 55.8 | Improved |
| add multi-seed R-CNN (MSR) supervision for the R-CNN head (ablation) | CBL · PASCAL VOC 2007 test | mAP: 56.0 -> 57.4 | Improved |
| use WET ensemble during inference (ablation) | CBL · PASCAL VOC 2007 test | mAP: 55.8 -> 56.0 | Improved |
| update WET with weighted EMA over all OICs and CLS instead of plain average (ablation) | WET · PASCAL VOC 2007 test | mAP: 56.1 -> 57.4 | Improved |
| update WET from the CLS branch alone rather than last OIC (ablation) | WET · PASCAL VOC 2007 test | mAP: 55.3 -> 54.5 | Hurt |
| linearly grow CRD overlap threshold instead of static 0.75 (ablation) | CRD · PASCAL VOC 2007 test | mAP: 55.8 -> 57.4 | Improved |
| linear decline of overlap threshold instead of growth (ablation) | CRD · PASCAL VOC 2007 test | mAP: 57.4 -> 56.2 | Hurt |
| build WET from whole structure including feature extractor rather than head only (ablation) | WET · PASCAL VOC 2007 test | mAP: 56.9 -> 57.4 | Improved |
| EMA rate 0.999 versus 0.99 or 0.9999 (ablation) | WET · PASCAL VOC 2007 test | mAP: 55.2 / 56.4 -> 57.4 | Improved |
| vary confidence exponent gamma from 0.2 to 1.0 (ablation) | MSR · PASCAL VOC 2007 test | mAP: 57.2 -> 57.4 | No clear effect |
| add WET score to inference averaging (ablation) | CBL · PASCAL VOC 2007 test | mAP: 57.2 -> 57.4 | No clear effect |
Closest papers by meaning
- W2F: A Weakly-Supervised to Fully-Supervised Framework for Object Detection (CVPR 2018 · similarity 0.90)
- C-MIDN: Coupled Multiple Instance Detection Network With Segmentation Guidance for Weakly Supervised Object Detection (ICCV 2019 · similarity 0.90)
- PCL: Proposal Cluster Learning for Weakly Supervised Object Detection (TPAMI 2018 · similarity 0.89)
- W2N: Switching from Weak Supervision to Noisy Supervision for Object Detection (ECCV 2022 · similarity 0.89)
- Object-Aware Instance Labeling for Weakly Supervised Object Detection (ICCV 2019 · similarity 0.89)
- Instance-Aware, Context-Focused, and Memory-Efficient Weakly Supervised Object Detection (CVPR 2020 · 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.
