PointRFT: Explicit Reinforcement Fine-tuning for Point Cloud Few-shot Learning

TL;DR

PointRFT introduces a reinforcement fine-tuning approach for point cloud models, significantly improving few-shot learning performance by leveraging specialized reward functions and hybrid training paradigms.

Contribution

This paper presents the first reinforcement fine-tuning method for point cloud models, enhancing few-shot learning and integrating with pretraining for state-of-the-art results.

Findings

Outperforms vanilla supervised fine-tuning in few-shot classification

Effectively stabilizes training with accuracy and dispersion rewards

Achieves state-of-the-art results in data-scarce scenarios

Abstract

Understanding spatial dynamics and semantics in point cloud is fundamental for comprehensive 3D comprehension. While reinforcement learning algorithms such as Group Relative Policy Optimization (GRPO) have recently achieved remarkable breakthroughs in large language models by incentivizing reasoning capabilities through strategic reward design, their potential remains largely unexplored in the 3D perception domain. This naturally raises a pivotal question: Can RL-based methods effectively empower 3D point cloud fine-tuning? In this paper, we propose PointRFT, the first reinforcement fine-tuning paradigm tailored specifically for point cloud representation learning. We select three prevalent 3D foundation models and devise specialized accuracy reward and dispersion reward functions to stabilize training and mitigate distribution shifts. Through comprehensive few-shot classification experiments comparing distinct training paradigms, we demonstrate that PointRFT consistently outperforms vanilla supervised fine-tuning (SFT) across diverse benchmarks. Furthermore, when organically integrated into a hybrid Pretraining-SFT-RFT paradigm, the representational capacity of point cloud foundation models is substantially unleashed, achieving state-of-the-art performance particularly under data-scarce scenarios.