SHTOcc: Effective 3D Occupancy Prediction with Sparse Head and Tail Voxels

TL;DR

SHTOcc introduces a novel sparse head-tail voxel construction method for 3D occupancy prediction, effectively addressing class imbalance and geometric distribution issues, leading to improved accuracy, efficiency, and reduced memory usage.

Contribution

The paper proposes SHTOcc, a new approach that balances key voxels and reduces bias in 3D occupancy prediction, enhancing performance over existing methods.

Findings

Reduces GPU memory by 42.2%

Increases inference speed by 58.6%

Improves accuracy by approximately 7%

Abstract

3D occupancy prediction has attracted much attention in the field of autonomous driving due to its powerful geometric perception and object recognition capabilities. However, existing methods have not explored the most essential distribution patterns of voxels, resulting in unsatisfactory results. This paper first explores the inter-class distribution and geometric distribution of voxels, thereby solving the long-tail problem caused by the inter-class distribution and the poor performance caused by the geometric distribution. Specifically, this paper proposes SHTOcc (Sparse Head-Tail Occupancy), which uses sparse head-tail voxel construction to accurately identify and balance key voxels in the head and tail classes, while using decoupled learning to reduce the model's bias towards the dominant (head) category and enhance the focus on the tail class. Experiments show that significant improvements have been made on multiple baselines: SHTOcc reduces GPU memory usage by 42.2%, increases inference speed by 58.6%, and improves accuracy by about 7%, verifying its effectiveness and efficiency. The code is available at https://github.com/ge95net/SHTOcc