Research map / Detection and segmentation / CNN features
Panoptic SegFormer: Delving Deeper Into Panoptic Segmentation With Transformers
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 |
|---|---|---|---|
| Add mask-wise merging to DETR (ablation) | DETR-R50 · COCO val | PQ: 43.4 -> 44.7 | Helped |
| Add multi-scale deformable attention (ablation) | DETR panoptic segmentation · COCO val | PQ: 43.4 -> 47.3 | Helped |
| Use the proposed mask decoder (ablation) | DETR-R50 with multi-scale deformable attention · COCO val | PQ: 47.3 -> 48.5 | Helped |
| Decouple thing and stuff queries rather than use joint matching (ablation) | Panoptic SegFormer-R50 · COCO val2017 | PQ st: 39.5 -> 42.4 | Helped |
| Add location decoder layers to learn thing location clues (ablation) | Panoptic SegFormer-R50 · COCO val | PQ: 47.0 -> 49.6 | Helped |
| Predict mask mass centers instead of bounding boxes (ablation) | Panoptic SegFormer location decoder · COCO val | PQ: 49.2 vs. 49.6 | No clear effect |
| Replace mask-wise merging with pixel-wise argmax (ablation) | Panoptic SegFormer-R50 · COCO val | PQ: 49.6 -> 48.4 | Hurt |
| Use both classification probability and segmentation quality in mask confidence (ablation) | Mask-wise merging · COCO val | PQ: 48.7 -> 49.6 | Helped |
| Use segmentation quality alone for mask confidence (ablation) | Mask-wise merging · COCO val | PQ: 49.6 -> 44.4 | Hurt |
| Increase mask decoder attention heads from 1 to 8 (ablation) | Panoptic SegFormer-R50 · COCO val | PQ: 49.2 -> 49.6 | Helped |
| Apply random crops during training (ablation) | Panoptic SegFormer instance segmentation model · COCO test-dev | AP seg: 40.4 -> 41.7 | Helped |
| Use the Panoptic SegFormer architecture instead of MaskFormer with Swin-L | Panoptic segmentation · COCO-C | Mean PQ: 41.7 -> 47.2 | Helped |
Closest papers by meaning
- SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (NeurIPS 2021 · similarity 0.91)
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.89)
- Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (CVPR 2020 · similarity 0.88)
- UPSNet: A Unified Panoptic Segmentation Network (CVPR 2019 · similarity 0.88)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.88)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.88)
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.