3D-RFT: Reinforcement Fine-Tuning for Video-based 3D Scene Understanding

TL;DR

This paper introduces 3D-RFT, a reinforcement fine-tuning framework that directly optimizes large language models for video-based 3D scene understanding tasks using verifiable reward functions, achieving state-of-the-art results.

Contribution

The paper presents the first application of reinforcement fine-tuning with verifiable rewards to video-based 3D perception, improving alignment with evaluation metrics over traditional supervised methods.

Findings

3D-RFT-4B outperforms larger models on 3D detection and reasoning tasks.

Reinforcement fine-tuning with task-specific rewards enhances model performance.

The framework demonstrates robustness and valuable training insights.

Abstract

Reinforcement Learning with Verifiable Rewards ( RLVR ) has emerged as a transformative paradigm for enhancing the reasoning capabilities of Large Language Models ( LLMs), yet its potential in 3D scene understanding remains under-explored. Existing approaches largely rely on Supervised Fine-Tuning ( SFT), where the token-level cross-entropy loss acts as an indirect proxy for optimization, leading to a misalignment between training objectives and task performances. To bridge this gap, we present Reinforcement Fine-Tuning for Video-based 3D Scene Understanding (3D-RFT ), the first framework to extend RLVR to video-based 3D perception and reasoning. 3D-RFT shifts the paradigm by directly optimizing the model towards evaluation metrics. 3D-RFT first activates 3D-aware Multi-modal Large Language Models ( MLLM s) via SFT, followed by reinforcement fine-tuning using Group Relative Policy Optimization ( GRPO) with strictly verifiable reward functions. We design task-specific reward functions directly from metrics like 3D IoU and F1-Score to provide more effective signals to guide model training. Extensive experiments demonstrate that 3D-RFT-4B achieves state-of-the-art performance on various video-based 3D scene understanding tasks. Notably, 3D-RFT-4B significantly outperforms larger models (e.g., VG LLM-8B) on 3D video detection, 3D visual grounding, and spatial reasoning benchmarks. We further reveal good properties of 3D-RFT such as robust efficacy, and valuable insights into training strategies and data impact. We hope 3D-RFT can serve as a robust and promising paradigm for future development of 3D scene understanding.