Enhancing Indoor Occupancy Prediction via Sparse Query-Based Multi-Level Consistent Knowledge Distillation

TL;DR

This work tackles the efficiency-accuracy trade-off in 3D indoor occupancy prediction by proposing DiScene, a sparse query-based framework that uses Multi-level Consistent Knowledge Distillation to transfer knowledge from a large teacher to a lightweight student. The method coordinates encoder-level feature alignment, query-level matching with Hungarian assignment, and prior- and anchor-level distillations, complemented by a Teacher-Guided Initialization policy to accelerate convergence. Empirical results on Occ-ScanNet show DiScene achieving real-time inference (23.2 FPS) without depth priors and substantial mIoU gains over strong baselines, with depth priors enabling state-of-the-art performance while keeping feasible speed. Additional validation on Occ3D-nuScenes and in-the-wild data demonstrates robustness and versatility, underscoring practical value for robotics and real-time scene understanding.

Abstract

Occupancy prediction provides critical geometric and semantic understanding for robotics but faces efficiency-accuracy trade-offs. Current dense methods suffer computational waste on empty voxels, while sparse query-based approaches lack robustness in diverse and complex indoor scenes. In this paper, we propose DiScene, a novel sparse query-based framework that leverages multi-level distillation to achieve efficient and robust occupancy prediction. In particular, our method incorporates two key innovations: (1) a Multi-level Consistent Knowledge Distillation strategy, which transfers hierarchical representations from large teacher models to lightweight students through coordinated alignment across four levels, including encoder-level feature alignment, query-level feature matching, prior-level spatial guidance, and anchor-level high-confidence knowledge transfer and (2) a Teacher-Guided Initialization policy, employing optimized parameter warm-up to accelerate model convergence. Validated on the Occ-Scannet benchmark, DiScene achieves 23.2 FPS without depth priors while outperforming our baseline method, OPUS, by 36.1% and even better than the depth-enhanced version, OPUS†. With depth integration, DiScene† attains new SOTA performance, surpassing EmbodiedOcc by 3.7% with 1.62$\times$ faster inference speed. Furthermore, experiments on the Occ3D-nuScenes benchmark and in-the-wild scenarios demonstrate the versatility of our approach in various environments. Code and models can be accessed at https://github.com/getterupper/DiScene.

Figures (7)

• Figure 1: We compare DiScene with existing indoor occupancy prediction methods in terms of speed and accuracy. All models are evaluated on the Occ-ScanNet yu2024monocular validation set and inference speeds are measured on one NVIDIA A800 GPU w/o TensorRT. The size of the circle represents the model’s size.
• Figure 2: (a) The illustration of our proposed knowledge distillation strategies. (b) The architecture of our primary framework. Best viewed in color.
• Figure 3: Qualitative results of occupancy prediction on the Occ-ScanNet validation set. Compared with existing methods, DiScene demonstrates superior geometric awareness and semantic comprehension, visually highlighted by red and yellow boxes respectively.
• Figure 4: Visualization of occupancy predictions and activated query distributions across non-distilled student, distilled student, and teacher models. Activated queries are highlighted in red, with higher density near ground-truth regions indicating superior performance. The size of the circle represents the distance of the query center: larger circles are closer to the camera. We adjust the opacity of certain figures for better illustration. Best viewed in color.
• Figure 5: Qualitative results of occupancy prediction on the Occ3D-nuScenes validation set. The boxes highlight finer local detail capture (Row 1) and enhanced scene structure reconstruction (Row 2) achieved by our distilled model.