Research map / Detection and segmentation / CNN features
Towards Deeply Unified Depth-aware Panoptic Segmentation with Bi-directional Guidance Learning
Junwen He, Yifan Wang, Lijun Wang, Huchuan Lu, Bin Luo, Jun-Yan He, Jin-Peng Lan, Yifeng Geng, Xuansong Xie
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| replace depth regression head with instance-wise depth decoder using unified queries (ablation) | DeepDPS (ResNet-50) · Cityscapes-DVPS | DPQ: 60.3 -> 62.1 | Improved |
| add backup query for blank mask regions (ablation) | DeepDPS (ResNet-50) · Cityscapes-DVPS | DPQ: 62.1 -> 62.4 | Improved |
| add semantic-to-depth guidance (ablation) | DeepDPS (ResNet-50) · Cityscapes-DVPS | DPQ: 62.4 -> 62.8 | Improved |
| add depth-to-semantic guidance (ablation) | DeepDPS (ResNet-50) · Cityscapes-DVPS | DPQ: 62.8 -> 63.0 | Improved |
| add bi-directional guidance under incomplete supervision (ablation) | DeepDPS (semi-supervised) · Cityscapes-DVPS incomplete labels | DPQ: 55.8 -> 57.8 | Improved |
| train with incomplete supervision labels instead of full supervision (ablation) | DeepDPS · Cityscapes-DVPS | DPQ: 58.5 -> 55.8 | Hurt |
| choose different guidance loss designs (SGT, DGS, ours) (ablation) | DeepDPS guidance learning · Cityscapes-DVPS | DPQ: 57.5 -> 58.0 (SGT), 58.2 (ours) | Improved |
| use learned latent representation for geometric query enhancement instead of single query (ablation) | DeepDPS · Cityscapes-DVPS | DPQ: 53.3 -> 58.5 | Improved |
| add a second latent representation for semantic features (ablation) | DeepDPS · Cityscapes-DVPS | DPQ: 58.5 -> 57.4 | Hurt |
| Ours (Swin-B) versus PolyphonicFormer (Swin-B) at lambda=0.5 | DeepDPS · Cityscapes-DVPS | DPQ (lambda=0.5): 70.6 vs 69.8; lambda=0.1 55.0 vs 50.2 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- PolyphonicFormer — PolyphonicFormer: Unified Query Learning for Depth-Aware Video Panoptic Segmentation (ECCV 2022) · better on 4 of 5 shared results
- PanopticDepth — PanopticDepth: A Unified Framework for Depth-Aware Panoptic Segmentation (CVPR 2022) · better on 3 of 3 shared results
- ViP-DeepLab — VIP-DeepLab: Learning Visual Perception With Depth-Aware Video Panoptic Segmentation (CVPR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PanopticDepth: A Unified Framework for Depth-Aware Panoptic Segmentation (CVPR 2022 · similarity 0.94)
- SDC-Depth: Semantic Divide-and-Conquer Network for Monocular Depth Estimation (CVPR 2020 · similarity 0.89)
- PolyphonicFormer: Unified Query Learning for Depth-Aware Video Panoptic Segmentation (ECCV 2022 · similarity 0.89)
- Real-Time Panoptic Segmentation From Dense Detections (CVPR 2020 · similarity 0.89)
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.89)
- 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.
