Research map / Detection and segmentation / Weakly supervised detection and segmentation
Tell Me What They're Holding: Weakly-Supervised Object Detection with Transferable Knowledge from Human-Object Interaction
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 |
|---|---|---|---|
| Transfer RRPN localization knowledge learned from HOI and use image, pose, and verb features | Faster R-CNN target-class detection · HICO-DET D_T(W) | mAP@.5: 9.57 -> 17.19 | Helped |
| Add the pose feature to image and verb features (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 16.34 -> 17.19 | Helped |
| Add the verb feature to image and pose features (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 10.07 -> 17.19 | Helped |
| Add the pose and verb features to image features (ablation) | RRPN · HICO-DET D_T(W) | Recall: 22.75 -> 28.64 | Helped |
| Train RRPN with pose features alone rather than image, pose, and verb features (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 17.19 -> 5.24 | Hurt |
| Remove image features, using pose and verb features only (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 17.19 -> 4.33 | Hurt |
| Freeze the shared image backbone during target-class training (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 17.19 -> 10.58 | Hurt |
| Set RRPN loss weight λ to 0 instead of 10 (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 1.61 -> 17.19 | Hurt |
| Set RRPN loss weight λ to 1 rather than 10 (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 9.27 -> 17.19 | Hurt |
| Set RRPN loss weight λ to 5 rather than 10 (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 14.38 -> 17.19 | Hurt |
| Set RRPN loss weight λ to 15 or 20 rather than 10 (ablation) | RRPN + Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 17.19 -> 15.85; 17.19 -> 15.75 | Hurt |
| Use attention threshold δ=0.05, 0.15, or 0.20 rather than 0.10 (ablation) | RRPN pseudo-box generation and Faster R-CNN · HICO-DET D_T(W) | mAP@.5: 9.41; 14.01; 13.24 vs 17.19 | Hurt |
Closest papers by meaning
- Weakly-supervised HOI Detection via Prior-guided Bi-level Representation Learning (ICLR 2023 · similarity 0.90)
- Track and Transfer: Watching Videos to Simulate Strong Human Supervision for Weakly-Supervised Object Detection (CVPR 2016 · similarity 0.89)
- Weakly Supervised Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.89)
- You Reap What You Sow: Using Videos to Generate High Precision Object Proposals for Weakly-Supervised Object Detection (CVPR 2019 · similarity 0.88)
- Learning to Detect Human-Object Interactions With Knowledge (CVPR 2019 · similarity 0.88)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.87)
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.