Research map / Detection and segmentation / Weakly supervised detection and segmentation
The Devil Is in the Points: Weakly Semi-Supervised Instance Segmentation via Point-Guided Mask Representation
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 |
|---|---|---|---|
| Use point-labeled images instead of unlabeled images alongside 5% fully labeled data | WSSIS with SOLOv2 · COCO test-dev, 5% fully labeled | AP: 25.6 -> 33.7 | Helped |
| Add point-labeled images alongside 7k fully labeled images | Instance segmentation on BDD100K · BDD100K validation | AP: 22.1 -> 27.9 | Helped |
| Select one FPN level by maximum confidence instead of using only P2 (ablation) | Point-guided pseudo-mask generation · COCO train5K | AP: 23.4 -> 28.6 | Helped |
| Select one FPN level by maximum confidence instead of using all levels (ablation) | Point-guided pseudo-mask generation · COCO train5K | AP: 10.3 -> 28.6 | Helped |
| Add MaskRefineNet to point-guided rough masks (ablation) | WSSIS · COCO validation, 10% fully labeled setting | AP: 32.2 -> 35.5 | Helped |
| Provide the rough mask as MaskRefineNet input instead of image alone (ablation) | MaskRefineNet · COCO train5K | AP: 14.8 -> 29.7 | Helped |
| Add point guidance to MaskRefineNet with the rough mask input (ablation) | MaskRefineNet · COCO train5K | AP: 29.7 -> 39.1 | Helped |
| Use random in-instance points instead of center points (ablation) | Point-guided pseudo-mask generation · COCO train5K | AP: 28.6 -> 28.8 | No clear effect |
| Use point labels rather than image-level labels as weak supervision (ablation) | WSSIS · COCO validation, 10% fully labeled setting | AP: 29.5 -> 32.2 | Helped |
| Increase MaskRefineNet input size from 128×128 to 256×256 (ablation) | MaskRefineNet · COCO validation | AP: 34.1 -> 35.5 | Helped |
| Increase MaskRefineNet input size from 256×256 to 384×384 (ablation) | MaskRefineNet · COCO validation | AP: 35.5 vs. 35.4 | No clear effect |
| Add iterative pseudo-label generation and retraining (ablation) | WSSIS · COCO validation, 1% fully labeled | AP: 23.9 -> 25.6 | Helped |
Closest papers by meaning
- Pointly-Supervised Instance Segmentation (CVPR 2022 · similarity 0.93)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.91)
- BoxTeacher: Exploring High-Quality Pseudo Labels for Weakly Supervised Instance Segmentation (CVPR 2023 · similarity 0.91)
- Weakly Supervised Instance Segmentation for Videos With Temporal Mask Consistency (CVPR 2021 · similarity 0.90)
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.89)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 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.