Research map / Detection and segmentation / 3D detection and driving
WayPOP: A Panoramic Open-Set Panoptic Tracking Benchmark
Rohit Mohan, Maximilian Luz, Swarnava Chowdhury, Yao Lu, Florian Drews, Thomas Nürnberg, Yakov Miron, Federico Tombari, Stefano Gasperini, Abhinav Valada
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces WayPOP, a panoramic open-set panoptic tracking benchmark for autonomous driving that evaluates detecting, segmenting, and tracking out-of-distribution objects across time and multiple camera views.
Problem
Existing panoptic tracking benchmarks assume a fixed semantic taxonomy and so cannot test whether panoramic multi-camera driving systems detect, segment, and track out-of-distribution objects across time and viewpoints.
Approach
A benchmark designed for open-set panoramic panoptic tracking, so that OOD objects must be detected, segmented and tracked across time and multiple cameras rather than only classified within a closed taxonomy.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use open-set methods instead of open-vocabulary Mask2Former-style methods for unknown objects | panoptic segmentation baselines · WayPOP-v1 VI-OSPS | Unknown PQ: FC-CLIP 0.2 -> PoDS 34.6 | Improved |
| use caption-supervised ODISE | open-vocabulary segmentation · WayPOP-v1 VI-OSPS | Unknown PQ: 30.9 vs <=1.0 for other OV methods | Improved |
| use appearance-based trackers (MASA, DEVA) instead of motion-based BoT-SORT | FV-OSPT · WayPOP-v1 | Unknown AQ: PoDS: 14.8 -> 20.7 (MASA) | Improved |
| extend from frontal-view to multi-view tracking | PoDS + DEVA · WayPOP-v1 | Unknown STQ vs wSTQ: 30.0 (FV) vs 37.1 (MV) different metrics | Mixed |
| use ground-truth masks (oracle) with DEVA | MV-OSPT · WayPOP-v1 | Known wAQ: 11.4 -> 43.5 | Improved |
| change ODISE unknown prompt from 'unknown object' to full vocabulary (ablation) | ODISE · WayPOP-v1 | Unknown PQ: 7.7 -> 30.9 | Improved |
| change ODISE unknown prompt (ablation) | ODISE · WayPOP-v1 | Known PQ: 24.0 to 24.4 | No clear effect |
| add lighting adaptation, occlusion handling and harmonization to insertion pipeline (ablation) | benchmark construction · WayPOP-v1 dev, RbA | Unknown PQ: 38.0 -> 29.9 | Hurt |
| add lighting adaptation, occlusion handling and harmonization (ablation) | benchmark construction · WayPOP-v1 | FID: 14.762 -> 9.745 | Improved |
| use Objaverse-XL assets instead of Objaverse v1.0 | open-set methods · WayPOP VI-OSPS | Unknown PQ: PoDS 34.6 -> 31.1 | No clear effect |
| use Objaverse-XL assets instead of v1.0 | ODISE · WayPOP VI-OSPS | Unknown PQ: 30.9 -> 23.6 | Hurt |
Experimental setup
Benchmarks: WayPOP
Closest papers by meaning
- LiDAR Prompted Spatio-Temporal Multi-View Stereo for Autonomous Driving (CVPR 2026 · similarity 0.88)
- ZOPP: A Framework of Zero-shot Offboard Panoptic Perception for Autonomous Driving (NeurIPS 2024 · similarity 0.87)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · similarity 0.86)
- OpenAD: Open-World Autonomous Driving Benchmark for 3D Object Detection (NeurIPS 2025 · similarity 0.86)
- Panacea: Panoramic and Controllable Video Generation for Autonomous Driving (CVPR 2024 · similarity 0.86)
- DriveGEN: Generalized and Robust 3D Detection in Driving via Controllable Text-to-Image Diffusion Generation (CVPR 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.
