Research map / Detection and segmentation / Weakly supervised detection and segmentation
Comprehensive Attention Self-Distillation for Weakly-Supervised Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 input-wise self-distillation over image transformations (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: 48.9 -> 54.1 | Helped |
| Add layer-wise self-distillation over feature blocks (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: 48.9 -> 52.3 | Helped |
| Combine input-wise and layer-wise CASD (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: 48.9 -> 55.3 | Helped |
| Add proposal score aggregation to input-wise CASD without inverted attention (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: 51.4 -> 52.6; 1.2% | Helped |
| Add inverted attention to input-wise CASD (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: +1.5% | Helped |
| Add the regression branch to IW+LW CASD (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: 55.3 -> 56.1; 0.8% | Helped |
| Add color augmentation to flip+scale augmentation (ablation) | CASD with regression branch · VOC 2007 test | mAP@0.5: 56.1 -> 56.8 | Helped |
| Add B1 to the B4+B3+B2 layer configuration (ablation) | Layer-wise CASD · VOC 2007 test | mAP@0.5: 52.3 -> 51.4 | Hurt |
| Use CASD attention distillation instead of prediction or attention consistency (ablation) | OICR WSOD with VGG16 · VOC 2007 test | mAP@0.5: 50.2 -> 52.6; 51.0 -> 52.6 | Helped |
| Replace average-pooling+sigmoid attention with Grad-CAM (ablation) | Layer-wise CASD · VOC 2007 test | mAP@0.5: 52.6 -> 52.0 | Hurt |
| Use different weights for IW and LW losses rather than one shared gamma (ablation) | CASD · VOC 2007 test | mAP@0.5: 55.3 -> 55.5; 0.2% | No clear effect |
| Use CASD rather than MIST(+Reg.) | WSOD with VGG16 · VOC 2007 test | mAP@0.5: 54.9 -> 56.8; 1.9% | Helped |
Closest papers by meaning
- Salvage of Supervision in Weakly Supervised Object Detection (CVPR 2022 · similarity 0.88)
- UWSOD: Toward Fully-Supervised-Level Capacity Weakly Supervised Object Detection (NeurIPS 2020 · similarity 0.88)
- Object Instance Mining for Weakly Supervised Object Detection (AAAI 2020 · similarity 0.88)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.86)
- Weakly Supervised Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.85)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 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.