Problem
Research questions and friction points this paper is trying to address.
3D Occupancy Prediction
Scene Flow Prediction
Spatiotemporal Aggregation
Autonomous Driving
SelectOccFlow: Selective Spatiotemporal Aggregation for 3D Occupancy and Scene Flow Prediction
🤖 AI Summary
This study addresses the unreliability of spatiotemporal aggregation in camera-based 3D occupancy and scene flow prediction by proposing a selective spatiotemporal aggregation framework. The framework comprises three core modules: Semantic-Guided Sampling (SGS) to resolve semantic incompatibility in features, State-Conditional Temporal Aggregation (SCTA) to correct misalignments in historical observations, and Extent-Aware Spatial Aggregation (ESA) to repair missing voxel structures. Together, these modules enable the reliable fusion of evidence across the image, temporal, and voxel domains. Evaluated on the OpenOcc dataset, the proposed method achieves a state-of-the-art OccScore of 44.9, representing a 4.2% improvement over prior approaches. Furthermore, under nuScenes-C corruption scenarios, it yields an average score increase of 11.1%, demonstrating significantly enhanced model robustness.
📝 Abstract
Comprehensive 3D scene understanding for autonomous driving requires modeling geometry, semantics, and motion. However, camera-based occupancy and scene flow prediction are sensitive to unreliable spatial and temporal aggregation, caused by semantically incompatible image features, misaligned historical observations, and incomplete voxel structures. To address this issue, we propose SelectOccFlow, a selective spatiotemporal aggregation framework that progressively refines contextual evidence across image, temporal, and voxel domains. To obtain semantically compatible image evidence, we design Semantic-Guided Sampling (SGS) to regulate feature sampling with semantic priors. Since reliable image evidence alone cannot resolve temporal inconsistency, we then present State-Conditioned Temporal Aggregation (SCTA) to selectively retrieve historical evidence according to voxel states. To further enhance the structural completeness of voxel representations, we introduce Extent-Aware Spatial Aggregation (ESA), which exploits directional structural support to refine foreground geometry. Experiments on OpenOcc demonstrate that SelectOccFlow achieves a state-of-the-art OccScore of 44.9, improving the previous best by +4.2%. It also maintains competitive occupancy performance on Occ3D-nus and improves the mean OccScore under nuScenes-C corruptions by +11.1%, demonstrating improved robustness to visual corruptions. The source code will be made publicly available at https://github.com/muchen1021/SelectOccFlow.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Selective Spatiotemporal Aggregation
Semantic-Guided Sampling
State-Conditioned Temporal Aggregation
Extent-Aware Spatial Aggregation
3D Occupancy Prediction
🔎 Similar Papers
No similar papers found.
