Research map / Detection and segmentation / CNN features
PolyphonicFormer: Unified Query Learning for Depth-Aware Video Panoptic Segmentation
Haobo Yuan, Xiangtai Li, Yibo Yang, Guangliang Cheng, Jing Zhang, Yunhai Tong, Lefei Zhang, Dacheng Tao
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 |
|---|---|---|---|
| add instance-level depth prediction (ablation) | PolyphonicFormer (Hybrid) · Cityscapes-DVPS (ablation, ResNet-50) | abs rel: 0.089 -> 0.080 | Improved |
| add depth supervision to panoptic query model (ablation) | Panoptic baseline · Cityscapes-DVPS (ablation, ResNet-50) | PQ: 63.7 -> 65.1 | Improved |
| replace dense depth baseline by unified query learning (ablation) | Depth baseline · Cityscapes-DVPS (ablation, ResNet-50) | abs rel: 0.084 -> 0.080 | Improved |
| add Query Linking (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | PQ: 64.8 -> 65.2 | Improved |
| add Query Linking (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | abs rel: 0.088 -> 0.080 | Improved |
| add Dense Depth Initialization (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | PQ: 63.2 -> 65.2 | Improved |
| add Dense Depth Initialization (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | abs rel: 0.094 -> 0.080 | Improved |
| use bidirectional query linking instead of panoptic-to-depth add (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | PQ: 65.2 -> 63.8 | Hurt |
| use self-attention linking instead of add (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | PQ: 65.2 -> 65.3 | No clear effect |
| raise depth loss weight from 0.1 to 10 (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | PQ: 65.4 -> 65.4 | No clear effect |
| raise depth loss weight from 0.1 to 10 (ablation) | PolyphonicFormer · Cityscapes-DVPS (ablation, ResNet-50) | abs rel: 0.101 -> 0.079 | Improved |
| use QuasiDense tracker instead of DeepSort (ablation) | PolyphonicFormer · SemKITTI-DVPS (tracking ablation) | DSTQ: 51.8 -> 63.6 | Improved |
Later papers that compared with this method
1 later accepted paper reports PolyphonicFormer as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Towards Deeply Unified Depth-aware Panoptic Segmentation with Bi-directional Guidance Learning (ICCV 2023 · better on 4 of 5 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViP-Deeplab — VIP-DeepLab: Learning Visual Perception With Depth-Aware Video Panoptic Segmentation (CVPR 2021) · better on 6 of 6 shared results
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 0.90)
- Towards Deeply Unified Depth-aware Panoptic Segmentation with Bi-directional Guidance Learning (ICCV 2023 · similarity 0.89)
- Object-Centric Representation Learning for Video Scene Understanding (TPAMI 2024 · similarity 0.89)
- VIP-DeepLab: Learning Visual Perception With Depth-Aware Video Panoptic Segmentation (CVPR 2021 · similarity 0.88)
- Video Panoptic Segmentation (CVPR 2020 · similarity 0.87)
- IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals (NeurIPS 2025 · similarity 0.85)
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.
