Bidirectional Knowledge Reconfiguration for Lightweight Point Cloud Analysis

TL;DR

This paper introduces bidirectional knowledge reconfiguration (BKR), a novel feature distillation method for lightweight point cloud analysis models that effectively transfers local and global features from a teacher model to a student model, improving performance on mobile and edge devices.

Contribution

The paper proposes BKR, a new bidirectional knowledge distillation approach with feature reconfiguration and a feature mover's distance loss to align features despite sampling misalignments.

Findings

BKR improves point cloud classification accuracy.

The method enhances segmentation performance.

Experiments show universality across tasks.

Abstract

Point cloud analysis faces computational system overhead, limiting its application on mobile or edge devices. Directly employing small models may result in a significant drop in performance since it is difficult for a small model to adequately capture local structure and global shape information simultaneously, which are essential clues for point cloud analysis. This paper explores feature distillation for lightweight point cloud models. To mitigate the semantic gap between the lightweight student and the cumbersome teacher, we propose bidirectional knowledge reconfiguration (BKR) to distill informative contextual knowledge from the teacher to the student. Specifically, a top-down knowledge reconfiguration and a bottom-up knowledge reconfiguration are developed to inherit diverse local structure information and consistent global shape knowledge from the teacher, respectively. However, due to the farthest point sampling in most point cloud models, the intermediate features between teacher and student are misaligned, deteriorating the feature distillation performance. To eliminate it, we propose a feature mover's distance (FMD) loss based on optimal transportation, which can measure the distance between unordered point cloud features effectively. Extensive experiments conducted on shape classification, part segmentation, and semantic segmentation benchmarks demonstrate the universality and superiority of our method.