Research map / Detection and segmentation / Open-set and few-shot detection
PolyFormer: Referring Image Segmentation As Sequential Polygon Generation
Jiang Liu, Hui Ding, Zhaowei Cai, Yuting Zhang, Ravi Kumar Satzoda, Vijay Mahadevan, R. Manmatha
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 |
|---|---|---|---|
| Use continuous coordinate regression instead of coordinate classification (ablation) | PolyFormer-B · RefCOCO+ RIS val | mIoU: 68.79 -> 70.65 | Improved |
| Use continuous coordinate regression instead of coordinate classification (ablation) | PolyFormer-B · RefCOCO+ REC val | Precision@0.5: 81.35 -> 83.73 | Improved |
| Order polygon vertices clockwise from the vertex closest to the image origin (ablation) | PolyFormer-B · RefCOCO RIS val | mIoU: 55.92 -> 68.35 | Improved |
| Add polygon augmentation with varied granularity (ablation) | PolyFormer-B with ordered polygons · RefCOCO RIS val | mIoU: 68.35 -> 72.07 | Improved |
| Add multi-task learning of RIS and REC (ablation) | PolyFormer-B with ordered, augmented polygons · RefCOCO RIS val | mIoU: 72.07 -> 75.14 | Improved |
| Add the <SEP> token to represent multiple polygons (ablation) | PolyFormer-B with ordered, augmented polygons and multi-task learning · RefCOCO RIS val | mIoU: 75.14 -> 75.96 | Improved |
| Use a 64×64 rather than 32×32 coordinate codebook (ablation) | PolyFormer-B · RefCOCO+ RIS val | mIoU: 70.15 -> 70.65 | Improved |
| Use a 64×64 rather than 128×128 coordinate codebook (ablation) | PolyFormer-B · RefCOCO+ RIS val | mIoU: 70.01 -> 70.65 | Improved |
| Use PolyFormer-L instead of PolyFormer-B with a larger visual backbone and deeper transformer | PolyFormer · RefCOCO+ RIS val | mIoU: 70.65 -> 72.15 | Improved |
| Transfer the image-trained model to video without fine-tuning | PolyFormer-L · Ref-DAVIS17 | J&F: 61.5 | Improved |
Closest papers by meaning
- Instance Segmentation With Mask-Supervised Polygonal Boundary Transformers (CVPR 2022 · similarity 0.87)
- PolyTransform: Deep Polygon Transformer for Instance Segmentation (CVPR 2020 · similarity 0.85)
- Locate Then Segment: A Strong Pipeline for Referring Image Segmentation (CVPR 2021 · similarity 0.85)
- Language As Queries for Referring Video Object Segmentation (CVPR 2022 · similarity 0.84)
- LQMFormer: Language-aware Query Mask Transformer for Referring Image Segmentation (CVPR 2024 · similarity 0.84)
- Improving Transformer Based Line Segment Detection with Matched Predicting and Re-ranking (AAAI 2025 · similarity 0.83)
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.
