Research map / Detection and segmentation / Weakly supervised detection and segmentation
Point2Mask: Point-supervised Panoptic Segmentation via Optimal Transport
Wentong Li, Yuqian Yuan, Song Wang, Jianke Zhu, Jianshu Li, Jian Liu, Lei Zhang
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 |
|---|---|---|---|
| replace minimum-cost pixel assignment with global optimal transport for pseudo-masks (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 51.9 -> 54.2 | Improved |
| add the high-level boundary map to the transport cost (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 50.6 -> 53.4 | Improved |
| add the low-level (Structured Edge) boundary map to the cost (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 50.6 -> 51.1 | Improved |
| use both boundary maps together with the semantic map (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 53.4 -> 53.8 | Improved |
| add LAB and RGB affinity losses to partial cross-entropy for semantic learning (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 40.4 -> 53.8 | Improved |
| use Nearest Centroid rather than Equal Division for unit number k (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 52.4 -> 53.8 | Improved |
| use Nearest Centroid rather than Nearest gt Point for unit number k (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 52.8 -> 53.8 | Improved |
| iterate centroid update 8 times instead of 1 (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 53.76 -> 54.24 | Improved |
| increase centroid iterations from 8 to 10 (ablation) | Point2Mask · Pascal VOC 2012 val | PQ: 54.24 -> 54.07 | No clear effect |
| pre-train on point-labeled COCO before training on VOC | Point2Mask · Pascal VOC 2012 val (ResNet-50) | PQ: 53.8 -> 60.7 | Improved |
| scale backbone from ResNet-50 to Swin-L | Point2Mask · COCO val2017 | PQ: 32.4 -> 37.0 | Improved |
Closest papers by meaning
- Pointly-Supervised Panoptic Segmentation (ECCV 2022 · similarity 0.93)
- Fully Data-Driven Pseudo Label Estimation for Pointly-Supervised Panoptic Segmentation (AAAI 2024 · similarity 0.92)
- Point-supervised Panoptic Segmentation via Estimating Pseudo Labels from Learnable Distance (ECCV 2024 · similarity 0.91)
- Fully Convolutional Networks for Panoptic Segmentation with Point-based Supervision (TPAMI 2022 · similarity 0.89)
- Beyond the Label Itself: Latent Labels Enhance Semi-supervised Point Cloud Panoptic Segmentation (AAAI 2024 · similarity 0.88)
- Scene-Centric Unsupervised Panoptic Segmentation (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.
