Research map / Detection and segmentation / CNN features
PolarMask++: Enhanced Polar Representation for Single-Shot Instance Segmentation and Beyond
Enze Xie, Wenhai Wang, Mingyu Ding, Ruimao Zhang, Ping Luo
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 Smooth-L1 loss with Polar IoU loss (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 25.1 -> 27.7 | Improved |
| replace original centerness with polar centerness (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 27.7 -> 29.1 | Improved |
| use soft polar centerness (F3) instead of polar centerness (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 29.1 -> 29.8 | Improved |
| replace FPN with Refined Feature Pyramid (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 29.8 -> 30.2 | Improved |
| replace FPN with PANet (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 29.8 -> 29.9 | No clear effect |
| add a bounding box branch (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 27.5 -> 27.7 | No clear effect |
| increase number of rays from 18 to 24 (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 26.2 -> 27.3 | Improved |
| increase number of rays from 36 to 72 (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 27.7 -> 27.6 | No clear effect |
| enlarge input image to 1920x1280 | PolarMask++ ResNeXt-101-DCN · COCO test-dev | mask AP: 37.2 -> 38.7 | Improved |
| use ResNeXt-101 instead of ResNet-101 backbone (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 31.6 -> 33.9 | Improved |
| increase input shorter side from 600 to 800 (ablation) | PolarMask++ · COCO minival (ResNet-50) | mask AP: 28.5 -> 30.2 | Improved |
| add deformable convolution to ResNet-50 backbone (ablation) | PolarMask++ · COCO minival (12 epochs) | mask AP: 30.2 -> 33.2 | Improved |
Later papers that compared with this method
1 later accepted paper reports PolarMask-600 as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- SOTR: Segmenting Objects With Transformers (ICCV 2021 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- PolarMask — PolarMask: Single Shot Instance Segmentation With Polar Representation (CVPR 2020) · better on 2 of 2 shared results
- TensorMask — TensorMask: A Foundation for Dense Object Segmentation (ICCV 2019) · better on 1 of 1 shared result
- TextDragon — TextDragon: An End-to-End Framework for Arbitrary Shaped Text Spotting (ICCV 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PolarMask: Single Shot Instance Segmentation With Polar Representation (CVPR 2020 · similarity 0.97)
- PolarNeXt: Rethink Instance Segmentation with Polar Representation (CVPR 2025 · similarity 0.92)
- SOLO: A Simple Framework for Instance Segmentation (TPAMI 2021 · similarity 0.88)
- SipMask: Spatial Information Preservation for Fast Image and Video Instance Segmentation (ECCV 2020 · similarity 0.88)
- CenterMask: Single Shot Instance Segmentation With Point Representation (CVPR 2020 · similarity 0.87)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 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.
