Achieving highly accurate and real-time 3D occupancy prediction from cameras is a critical requirement for the safe and practical deployment of autonomous vehicles. While this shift to sparse 3D representations solves the encoding bottleneck, it creates a new challenge for the decoder: how to efficiently aggregate information from a sparse, non-uniformly distributed set of voxel features without resorting to computationally prohibitive dense attention. In this paper, we propose a novel Prototype-based Sparse Transformer Decoder that replaces this costly interaction with an efficient, two-stage process of guided feature selection and focused aggregation. Our core idea is to make the decoder's attention prototype-guided. We achieve this through a sparse prototype selection mechanism, where each query adaptively identifies a compact set of the most salient voxel features, termed prototypes, for focused feature aggregation. To ensure this dynamic selection is stable and effective, we introduce a complementary denoising paradigm. This approach leverages ground-truth masks to provide explicit guidance, guaranteeing a consistent query-prototype association across decoder layers. Our model, dubbed SPOT-Occ, outperforms previous methods with a significant margin in speed while also improving accuracy.
Occupancy Prediction on OpenOccupancy validation set:
Semantic Scene Completion on SemanticKITTI validation set:
We provide the pretrained weights on SemanticKITTI and nuScenes datasets.
| Model | Dataset | Backbone | SSC mIoU | Model Weight | Training Log | Inference Log |
|---|---|---|---|---|---|---|
| SparseOcc (Baseline) | nuScenes | ResNet50 | 13.2 | Link | Link | Link |
| SpotOcc (Ours) | nuScenes | ResNet50 | 13.7 | Link | Link | Link |
| SparseOcc (Baseline) | SemanticKITTI | EfficientNetB7 | 12.2 | Link | Link | Link |
| SpotOcc (Ours) | SemanticKITTI | EfficientNetB7 | 13.3 | Link | Link | Link |
If you find this work useful, please consider citing:
@article{spotocc2026,
title={SPOT-Occ: Sparse Prototype-guided Transformer for Camera-based 3D Occupancy Prediction},
author={Chen, Suzeyu and Li, Leheng and Chen, Ying-Cong},
journal={arXiv preprint arXiv:2602.04240},
year={2026}
}This project is developed based on the following open-sourced projects: BEVDet, BEVFormer, Mask2Former, OccFormer, OpenOccupancy, SparseOcc. Thanks for their excellent work.