Research map / Detection and segmentation / CNN features
Unifying Training and Inference for Panoptic Segmentation
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 dense instance affinity and panoptic matching loss (ablation) | Box-only panoptic segmentation model · Cityscapes val | PQ: +4.4 PQ | Helped |
| Add dense instance affinity and panoptic matching loss (ablation) | Mask-based panoptic model · Cityscapes val | PQ: 1.8 rise in overall PQ | Helped |
| Train end-to-end with panoptic matching loss (ablation) | Mask-based model evaluated with heuristic merger · Cityscapes val | PQ: 1.3 increase in PQ | Helped |
| Use direct argmax instead of heuristic merger (ablation) | End-to-end model with affinity · Cityscapes val | PQ: 0.8 PQ | Helped |
| Multiply semantic probabilities by mask scores rather than summing them (ablation) | Dynamic potential head · Cityscapes | PQ: 1.1 PQ lead | Helped |
| Sum semantic probabilities and mask scores rather than multiplying them (ablation) | Dynamic potential head · COCO | PQ: 0.7 PQ improvement | Helped |
| Use on-the-fly predicted detections instead of ground-truth detections during training (ablation) | Mask-free panoptic model · Cityscapes val | PQ: 0.4 increase in PQ | Helped |
| Use on-the-fly predicted detections instead of ground-truth detections during training (ablation) | Mask-free panoptic model · Cityscapes val | AP_box: +1.3 | Helped |
| Add a direct contrastive loss on affinity features (ablation) | Dense instance affinity head · Panoptic segmentation | PQ | No clear effect |
Closest papers by meaning
- Fully Convolutional Networks for Panoptic Segmentation (CVPR 2021 · similarity 0.92)
- Real-Time Panoptic Segmentation From Dense Detections (CVPR 2020 · similarity 0.92)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.92)
- Panoptic, Instance and Semantic Relations: A Relational Context Encoder To Enhance Panoptic Segmentation (CVPR 2022 · similarity 0.92)
- Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (CVPR 2020 · similarity 0.91)
- Bidirectional Graph Reasoning Network for Panoptic Segmentation (CVPR 2020 · similarity 0.91)
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.