MergeOcc: Bridge the Domain Gap between Different LiDARs for Robust Occupancy Prediction

TL;DR

This work tackles the domain gap that arises when applying LiDAR-based 3D occupancy prediction across heterogeneous LiDAR types. It introduces MergeOcc, a MDT-enabled framework that combines a geometric realignment module with a semantic label mapping module to train a single model on multiple datasets with distinct LiDARs. Empirical results on OpenOccupancy-nuScenes and SemanticKITTI show substantial cross-LiDAR improvements over single-dataset baselines and state-of-the-art methods, validating the approach and its scalability. The work highlights the practical impact of leveraging broader, diverse data to reduce deployment risk and labeling costs, while noting limitations in handling unseen LiDARs and outlining directions for future generalization and code release.

Abstract

LiDAR-based 3D occupancy prediction evolved rapidly alongside the emergence of large datasets. Nevertheless, the potential of existing diverse datasets remains underutilized as they kick in individually. Models trained on a specific dataset often suffer considerable performance degradation when deployed to real-world scenarios or datasets involving disparate LiDARs. This paper aims to develop a generalized model called MergeOcc, to simultaneously handle different LiDARs by leveraging multiple datasets. The gaps among LiDAR datasets primarily manifest in geometric disparities and semantic inconsistencies. Thus, MergeOcc incorporates a novel model featuring a geometric realignment module and a semantic label mapping module to enable multiple datasets training (MDT). The effectiveness of MergeOcc is validated through experiments on two prominent datasets for autonomous vehicles: OpenOccupancy-nuScenes and SemanticKITTI. The results demonstrate its enhanced robustness and remarkable performance across both types of LiDARs, outperforming several SOTA multi-modality methods. Notably, despite using an identical model architecture and hyper-parameter set, MergeOcc can significantly surpass the baseline due to its exposure to more diverse data. MergeOcc is considered the first cross-dataset 3D occupancy prediction pipeline that effectively bridges the domain gap for seamless deployment across heterogeneous platforms.

Figures (6)

• Figure 1: Challenges of cross-LiDAR issue: 1) Only OO-nu and Only SK refer to the baseline model trained on each individual dataset. 2) Direct Merging represents training the same model on the simply merged dataset. 3) MergeOcc denotes the proposed method trained on these two datasets.
• Figure 2: The framework of MergeOcc including: 1) point range alignment, 2) cylindrical voxelization, 3) shared 3D backbone with dataset-specific normalization, 4) dataset-specific occupancy heads, 5) semantic label mapping, and 6) coarse to fine stage.
• Figure 3: Comparison of previous label space intersection and our unified label space
• Figure 4: Visualizations of occupancy prediction results. The upper half illustrates the outcomes of nuScenes, while the lower half showcases the SemanticKITTI results .
• Figure 5: Visualizations of occupancy prediction results on OpenOccupancy-nuScenes.