Research map / Detection and segmentation / Weakly supervised detection and segmentation
Pointly-Supervised Instance Segmentation
Check 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 |
|---|---|---|---|
| Use bounding boxes plus 10 labeled random points per object instead of full masks | Mask R-CNN · COCO val2017 | AP: 37.2 -> 36.1 | Hurt |
| Use bounding boxes plus 10 labeled random points per object instead of full masks | Mask R-CNN · PASCAL VOC val | AP50: 66.3 -> 64.2 | Hurt |
| Use bounding boxes plus 10 labeled random points per object instead of full masks | Mask R-CNN · Cityscapes val | AP: 32.7 -> 30.7 | Hurt |
| Use bounding boxes plus 10 labeled random points per object instead of full masks | Mask R-CNN · LVISv1.0 val | AP: 22.8 -> 21.5 | Hurt |
| Replace PointRend's coarse mask head with an instance-conditioned implicit point head | PointRend on COCO val2017 with full masks · COCO val2017 | AP: 38.3 -> 38.5 (+0.2) | Helped |
| Replace PointRend's coarse mask head with an instance-conditioned implicit point head | PointRend on COCO val2017 with 10-point supervision · COCO val2017 | AP: 35.7 -> 36.9 (+1.2) | Helped |
| Add point-based augmentation by randomly subsampling half the labeled points per iteration (ablation) | Mask R-CNN with ResNeXt-101-32×8-FPN · COCO val2017 | AP: 38.1 -> 38.5 (+0.4 AP) | Helped |
| Add point-based augmentation by randomly subsampling half the labeled points per iteration (ablation) | Implicit PointRend · COCO val2017 | AP: 36.0 -> 36.9 (+0.9) | Helped |
| Flip labels for 5% of randomly selected points (ablation) | Mask R-CNN with 10-point supervision · COCO val2017 | AP: 36.1 -> 35.9 (0.2 AP worse) | Hurt |
| Flip labels for 5% of points nearest object boundaries (ablation) | Mask R-CNN with 10-point supervision · COCO val2017 | AP: 36.1 -> 35.7 (0.4 AP) | Hurt |
| Use points biased toward object boundaries rather than uniform random points (ablation) | Mask R-CNN with point supervision · COCO | AP: 36.1 AP -> 35.5 AP -> 27.5 AP | Hurt |
| Add coordinates relative to the bounding box to the point head (ablation) | Implicit PointRend · COCO | AP with 10-point supervision: 35.2 -> 36.6 (+1.4) | Helped |
Closest papers by meaning
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.91)
- Sketchy Bounding-box Supervision for 3D Instance Segmentation (CVPR 2025 · similarity 0.89)
- P2Seg: Pointly-supervised Segmentation via Mutual Distillation (ICLR 2024 · similarity 0.88)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.88)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.88)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (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.