Research map / Detection and segmentation / CNN features
Real-Time Panoptic Segmentation From Dense Detections
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 |
|---|---|---|---|
| Use separate localization and semantics towers instead of one shared tower (ablation) | Panoptic head · Cityscapes val | PQ: 56.8 -> 57.1 | Helped |
| Add levelness prediction to the two-tower model (ablation) | Panoptic head · Cityscapes val | PQ: 57.1 -> 58.1 | Helped |
| Add explicit mask loss to the model with two towers and levelness (ablation) | Panoptic head · Cityscapes val | PQ: 58.1 -> 58.8 | Helped |
| Train without instance-mask labels and omit mask loss, using box and semantic labels (ablation) | Ours · Cityscapes val | PQ: 58.8 -> 55.7 | Hurt |
| Use a shallower ResNet-18-FPN instead of ResNet-50-FPN (ablation) | Ours · Cityscapes full image size (1024 × 2048 px) | PQ: 58.8 -> 55.5 | Hurt |
| Use a shallower ResNet-34-FPN instead of ResNet-50-FPN (ablation) | Ours · Cityscapes full image size (1024 × 2048 px) | PQ: 58.8 -> 56.7 | Hurt |
| Use half-resolution input instead of full-resolution input (ablation) | Ours with ResNet-50-FPN · Cityscapes | PQ: 58.8 -> 49.5 | Hurt |
| Use the proposed single-stage model instead of UPSNet | Panoptic segmentation · COCO val | inference time: $110\text{ms}^*$ -> 63ms | Helped |
| Use the proposed single-stage model instead of the strongest reported two-stage method | Panoptic segmentation · COCO val | PQ: 42.5 -> 37.1 | Hurt |
| Use the proposed single-stage model instead of the strongest reported two-stage method | Panoptic segmentation · Cityscapes val | PQ: 60.2 -> 58.8 | Hurt |
Closest papers by meaning
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.92)
- PanopticDepth: A Unified Framework for Depth-Aware Panoptic Segmentation (CVPR 2022 · similarity 0.91)
- LPSNet: A Lightweight Solution for Fast Panoptic Segmentation (CVPR 2021 · similarity 0.91)
- Panoptic Segmentation (CVPR 2019 · similarity 0.91)
- You Only Segment Once: Towards Real-Time Panoptic Segmentation (CVPR 2023 · similarity 0.91)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.90)
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.