Research map / Detection and segmentation / Weakly supervised detection and segmentation
Active Pointly-Supervised Instance Segmentation
Chufeng Tang, Lingxi Xie, Gang Zhang, Xiaopeng Zhang, Qi Tian, Xiaolin Hu
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 |
|---|---|---|---|
| select points by entropy of anchor-averaged predictions instead of randomly | CondInst R-50 APIS · MS-COCO val2017 | mask mAP: 31.97 -> 32.88 -> 33.93 -> 34.49 vs random 32.32 -> 33.01 -> 33.69 | Improved |
| choose most certain points (lowest entropy) instead of most uncertain (ablation) | Entropy strategy · MS-COCO val2017 | mask mAP: 32.16, 32.50, 32.95 vs random 32.32, 33.01, 33.69 | Hurt |
| choose points with maximum error instead of entropy (ablation) | Entropy strategy · MS-COCO val2017 | mask mAP: 7.95, 12.45, 14.24 | Hurt |
| use entropy metric instead of variance on the prediction set (ablation) | APIS point selection · MS-COCO val2017 | mask mAP | Improved |
| build prediction set from multiple anchors instead of multiple models or scales (ablation) | Entropy strategy · MS-COCO val2017 | mask mAP | Improved |
| longer 90k fine-tuning schedule instead of shorter ones (ablation) | APIS Entropy · MS-COCO val2017 | mAP gap over random: +1.77% at the third step | Improved |
| transfer points acquired from CondInst R-50 to other models instead of selecting for them (ablation) | CondInst R-101 and BoxInst · MS-COCO val2017 | mask mAP | Mixed |
| use BoxInst as initial model, removing need of initial random points (ablation) | APIS · MS-COCO val2017 | mask mAP: 0.47% mAP | Improved |
| select by mean entropy of box points for mask labeling instead of random (AFIS) (ablation) | AFIS · MS-COCO val2017 | mask mAP | Hurt |
| select instances with lowest detection loss for mask labeling (AFIS) (ablation) | AFIS · MS-COCO val2017 | mask mAP | Improved |
| actively acquired points instead of random on object scales | Entropy strategy · MS-COCO val2017 (small objects) | AP_S: -0.30, +0.37, +0.28 | Mixed |
| actively acquired points instead of random on large instances | Entropy strategy · MS-COCO val2017 (large objects) | AP_L: +1.17, +1.53, +1.33 | Improved |
Closest papers by meaning
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.89)
- P2Seg: Pointly-supervised Segmentation via Mutual Distillation (ICLR 2024 · similarity 0.88)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.87)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.86)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.86)
- Density Matters: Improved Core-Set for Active Domain Adaptive Segmentation (AAAI 2024 · similarity 0.86)
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.
