Welcome to the Large-scale 3D Semantic Occupancy Prediction Challenge organized at
WACV 2026 Workshop / Challenge Info: TBA (date, venue, online link)
Paper
- Paper & Dataset Curation (arXiv): https://arxiv.org/abs/2510.22973
- Project / Code (UniScene v2): https://github.com/Arlo0o/UniScene-Unified-Occupancy-centric-Driving-Scene-Generation/tree/v2
Video
Coming soon…
📌 Overview
Semantic occupancy prediction is a cornerstone of 3D scene understanding in autonomous driving, providing dense, voxel-level semantic and geometric representations of dynamic environments. Despite recent advances, large-scale, high-resolution occupancy datasets remain scarce, limiting the development and evaluation of robust models.
This challenge introduces NuPlan-Occ, the largest publicly available semantic occupancy dataset to date, featuring 3.6 million frames with high-resolution voxel annotations (400 × 400 × 32). Derived from the widely adopted NuPlan benchmark, NuPlan-Occ enables scalable training and evaluation of both discriminative and generative models for 3D scene understanding.
We invite researchers to develop and submit models for 3D semantic occupancy prediction, with the goal of advancing state-of-the-art performance in accuracy, scalability, and generalization.
🎯 Task
Participants are tasked with predicting 3D semantic occupancy grids from multi-view camera images.
The occupancy grid is defined over a predefined spatial volume with 9 semantic classes including:
vehicle, pedestrian, bicycle, traffic_cone, barrier, czone_sign, generic_object, background, empty
Two tracks are supported:
- Discriminative Track: Predict occupancy from input images.
- Generative Track: Forecast occupancy sequences or complete partial scenes.
📊 Dataset: NuPlan-Occ
- Source: Built from the NuPlan dataset, with dense 3D semantic annotations.
- Scale: 19K scenes, 3.6M frames.
- Resolution: (400 × 400 × 32) voxels.
- Annotations: 10 semantic classes, with foreground/background separation.
- Modalities: Multi-view RGB images, LiDAR point clouds, BEV maps, and semantic occupancy grids.
Access
- HuggingFace: https://huggingface.co/datasets/Arlolo0/Nuplan-Occupancy
- GitHub: https://github.com/Arlo0o/UniScene-Unified-Occupancy-centric-Driving-Scene-Generation/tree/v2
- Paper & Dataset Curation: https://arxiv.org/abs/2510.22973
⚙️ Evaluation
Primary Metric: mean Intersection-over-Union (mIoU) across all semantic classes.
Secondary Metrics:
- Per-class IoU
- Precision & Recall
Evaluation code is released in the GitHub repository under the SOP/ directory:
https://github.com/Arlo0o/UniScene-Unified-Occupancy-centric-Driving-Scene-Generation/tree/v2/SOP/monoscene#3-evaluation
📝 Evaluation Guidelines
- Validation Phase: Participants can evaluate on a public validation set via our evaluation script.
- Reproducibility: Submitted code must include a README with training and inference instructions.
- Paper Submission: Top-performing teams will be invited to submit a short paper (4–6 pages) to the WACV 2026 workshop proceedings.
🏆 Baseline & Benchmark
We provide a reproduced baseline using MonoScene trained on NuPlan-Occ miniset:
| Metric | Value |
|---|---|
| Precision | 48.99 |
| Recall | 42.54 |
| IoU | 29.49 |
| mIoU | 9.36 |
Per-class IoU:
background: 29.0124, vehicle: 17.6694, bicycle: 0.4056, pedestrian: 6.4708, traffic_cone: 1.8878, barrier: 2.8173, czone_sign: 2.7924, generic_object: 13.8316
Baseline Code & Pretrained Model:
Released in the GitHub repository under the SOP/ directory.
https://github.com/Arlo0o/UniScene-Unified-Occupancy-centric-Driving-Scene-Generation/tree/v2/SOP/monoscene
🏅 Awards & Recognition
- 1st, 2nd, 3rd Place certificates and awards (sponsors TBD).
- Best Paper Award for the most innovative method.
- Top teams will be invited to present at the WACV 2026 workshop.
- Results will be published on the challenge website.
📚 Recommended Readings & Citations
Participants are encouraged to cite the following works:
```bibtex @inproceedings{li2025uniscene, title={Uniscene: Unified occupancy-centric driving scene generation}, author={Li, Bohan and Guo, Jiazhe and Liu, Hongsi and Zou, Yin…u and Tan, Feiyang and Zhang, Chi and Wang, Tiancai and others}, booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference}, pages={11971–11981}, year={2025} }
@article{li2025scaling, title={Scaling Up Occupancy-centric Driving Scene Generation: Dataset and Method}, author={Li, Bohan and Jin, Xin and Zhu, Hu and Liu, Hongsi and… Kaiwen and Ma, Chao and Jin, Yueming and Zhao, Hao and others}, journal={arXiv preprint arXiv:2510.22973}, year={2025} }