LighTROcc: Lightweight 4D Occupancy Forecasting via Instance-Centric 3D Gaussians

2026-10-07

AI Summary

This study addresses the prohibitive computational overhead and lack of instance consistency inherent in dense voxel representations, which hinder efficient future 3D occupancy prediction. To overcome these limitations, this work proposes a lightweight, instance-centric framework for 4D occupancy forecasting. The method introduces a compact instance representation based on attention-guided forward lifting, utilizing learned queries to localize objects. Furthermore, it employs anisotropic 3D Gaussian mixture models to represent and propagate displacements, enabling continuous spatiotemporal occupancy prediction within a single forward pass. Evaluated on the nuScenes dataset, the proposed approach preserves voxel-level fidelity while surpassing existing baselines in instance-level predictive accuracy. Consequently, this framework achieves a significant balance between forecasting precision and computational efficiency, offering a scalable solution for dynamic 3D scene understanding.

Abstract

Forecasting future 3D occupancy from surround-view cameras is essential for autonomous driving, yet existing approaches rely on dense voxel or bird's-eye-view representations whose cost grows rapidly with spatial resolution and prediction horizon. Because these representations do not explicitly maintain object identities, they also struggle to preserve instance consistency over time. We present LighTROcc, a lightweight instance-centric framework that represents movable objects with a compact set of learned queries and predicts present and future occupancy in a single forward pass. LighTROcc localizes each query through attention-guided forward lifting, combining image-space cross-attention, query-specific depth, and camera geometry to estimate its 3D center. Each instance is modeled as a mixture of anisotropic 3D Gaussians and propagated across future steps using predicted displacements, producing continuous, temporally consistent occupancy forecasts. Experiments on nuScenes and supplemented nuScenes-Occupancy show that LighTROcc outperforms the evaluated dense and instance-wise baselines in instance-level forecasting accuracy while maintaining strong voxel-level occupancy quality. Across different model configurations, LighTROcc achieves a favorable balance between forecasting accuracy and computational efficiency, demonstrating the potential of compact instance-centric modeling for camera-based 4D occupancy forecasting.

Problem

• 4D occupancy forecasting
• autonomous driving
• computational efficiency
• instance consistency
• surround-view cameras

Innovation

• 4D Occupancy Forecasting
• Instance-Centric
• 3D Gaussians
• Lightweight Framework
• Autonomous Driving