Research map / Detection and segmentation / CNN features
UPSNet: A Unified Panoptic Segmentation Network
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 |
|---|---|---|---|
| Train the panoptic head instead of using it only as post-processing (ablation) | UPSNet on COCO · COCO | PQ: 41.2 -> 41.6 | Helped |
| Add semantic-aware instance class assignment (ablation) | UPSNet on COCO · COCO | PQ: 41.6 -> 41.7 | Helped |
| Balance the loss weights (ablation) | UPSNet on COCO · COCO | PQ: 41.7 -> 42.3 | Helped |
| Add semantic-head RoI loss (ablation) | UPSNet on COCO · COCO | PQ: 42.3 -> 42.3 | No clear effect |
| Add unknown-class prediction (ablation) | UPSNet on COCO · COCO | PQ: 42.3 -> 42.5 | Helped |
| Replace predicted boxes with ground-truth boxes (ablation) | UPSNet on COCO · COCO | PQ: 42.5 -> 46.7 | Helped |
| Replace predicted instance class assignments with ground truth, using GT boxes (ablation) | UPSNet on COCO · COCO | PQ: 46.7 -> 53.0 | Helped |
| Replace predicted semantic segmentation with ground truth (ablation) | UPSNet on COCO · COCO | PQ: 53.0 -> 72.0 | Helped |
| Use UPSNet instead of MR-CNN-PSP | Panoptic segmentation · COCO val2017 | PQ: 41.8 -> 42.5 | Helped |
| Use UPSNet instead of MR-CNN-PSP | Panoptic segmentation · Cityscapes val | PQ: 58.0 -> 59.3 | Helped |
| Use UPSNet instead of MR-CNN-PSP | Panoptic segmentation · Our Dataset test | PQ: 45.5 -> 47.1 | Helped |
| Use multi-scale testing | UPSNet on COCO · COCO val2017 | PQ: 42.5 -> 43.0 | Helped |
Closest papers by meaning
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.91)
- LPSNet: A Lightweight Solution for Fast Panoptic Segmentation (CVPR 2021 · similarity 0.90)
- Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (CVPR 2020 · similarity 0.90)
- Attention-Guided Unified Network for Panoptic Segmentation (CVPR 2019 · similarity 0.90)
- Fully Convolutional Networks for Panoptic Segmentation (CVPR 2021 · similarity 0.90)
- 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.