Research map / Detection and segmentation / Weakly supervised detection and segmentation
Detecting 11K Classes: Large Scale Object Detection Without Fine-Grained Bounding Boxes
Hao Yang, Hao Wu, Hao Chen
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 |
|---|---|---|---|
| train jointly with fine-grained weakly-labeled data (naive two-stream) | SNIPER coarse-grained detector · OpenImages V4 val | mAP-CG: 45.7 -> 52.5 | Improved |
| add soft-attention proposal re-ranking (ablation) | naive two-stream · OpenImages V4 val | mAP-FG: 34.0 -> 49.2 | Improved |
| add soft-attention proposal re-ranking (ablation) | naive two-stream · OpenImages V4 val | mAP-CG: 52.5 -> 52.9 | No clear effect |
| add single-level coarse-grained memory (ablation) | +Soft-Attention · OpenImages V4 val | mAP-FG: 49.2 -> 49.6 | No clear effect |
| add single-level coarse-grained memory (ablation) | +Soft-Attention · OpenImages V4 val | mAP-CG: 52.9 -> 52.2 | Hurt |
| replace single-level memory with dual-level memory with foreground attention pooling (ablation) | +CG-Memory · OpenImages V4 val | mAP-FG: 49.6 -> 51.9 | Improved |
| replace single-level memory with dual-level memory with foreground attention pooling (ablation) | +CG-Memory · OpenImages V4 val | mAP-CG: 52.2 -> 53.5 | Improved |
| add fine-grained weak stream with shared backbone vs pure weakly-supervised fine-tuning (ablation) | naive two-stream · OpenImages V4 val | mAP-FG: 20.2 -> 34.0 | Improved |
| add fine-grained weakly labeled 3K classes to coarse detector | SNIPER coarse-grained detector · ImageNet | mAP-CG: 54.0 -> 50.7 | Hurt |
| scale fine-grained weak classes from 3K to 11K | Ours · ImageNet | mAP-CG: 50.7 -> 49.1 | Hurt |
| use weak-label semi-supervised model instead of full-supervision for fine-grained | Ours-3K vs SNIPER-FG-Fully · ImageNet | mAP-FG: 41.6 -> 35.1 | Hurt |
| use weak-label model instead of full-supervision at stricter IoU | Ours vs SNIPER-FG-Fully · ImageNet | mAP@0.5:0.95 FG: 22 -> 25 | Improved |
Closest papers by meaning
- Large Scale Semi-Supervised Object Detection Using Visual and Semantic Knowledge Transfer (CVPR 2016 · similarity 0.89)
- PreDet: Large-Scale Weakly Supervised Pre-Training for Detection (ICCV 2021 · similarity 0.89)
- Open-Vocabulary Object Detection Using Captions (CVPR 2021 · similarity 0.89)
- Visual and Semantic Knowledge Transfer for Large Scale Semi-Supervised Object Detection (TPAMI 2017 · similarity 0.89)
- End-to-End Semi-Supervised Object Detection With Soft Teacher (ICCV 2021 · similarity 0.89)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · 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.
