Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Xing, Zhenghao, Hu, Xiaowei, Fu, Chi-Wing, Wang, Wenhai, Dai, Jifeng, Heng, Pheng-Ann
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2505.04623
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912438633365504
author Xing, Zhenghao
Hu, Xiaowei
Fu, Chi-Wing
Wang, Wenhai
Dai, Jifeng
Heng, Pheng-Ann
author_facet Xing, Zhenghao
Hu, Xiaowei
Fu, Chi-Wing
Wang, Wenhai
Dai, Jifeng
Heng, Pheng-Ann
contents Multimodal large language models (MLLMs) have advanced perception across text, vision, and audio, yet they often struggle with structured cross-modal reasoning, particularly when integrating audio and visual signals. We introduce EchoInk-R1, a reinforcement learning framework that enhances such reasoning in MLLMs. Built upon the Qwen2.5-Omni-7B foundation and optimized with Group Relative Policy Optimization (GRPO), EchoInk-R1 tackles multiple-choice question answering over synchronized audio-image pairs. To enable this, we curate AVQA-R1-6K, a dataset pairing such audio-image inputs with multiple-choice questions derived from OmniInstruct-v1. EchoInk-R1-7B achieves 85.77% accuracy on the validation set, outperforming the base model, which scores 80.53%, using only 562 reinforcement learning steps. Beyond accuracy, EchoInk-R1 demonstrates reflective reasoning by revisiting initial interpretations and refining responses when facing ambiguous multimodal inputs. These results suggest that lightweight reinforcement learning fine-tuning enhances cross-modal reasoning in MLLMs. EchoInk-R1 is the first framework to unify audio, visual, and textual modalities for general open-world reasoning via reinforcement learning. Code and data are publicly released to facilitate further research.
format Preprint
id arxiv_https___arxiv_org_abs_2505_04623
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle EchoInk-R1: Exploring Audio-Visual Reasoning in Multimodal LLMs via Reinforcement Learning
Xing, Zhenghao
Hu, Xiaowei
Fu, Chi-Wing
Wang, Wenhai
Dai, Jifeng
Heng, Pheng-Ann
Computer Vision and Pattern Recognition
Audio and Speech Processing
Multimodal large language models (MLLMs) have advanced perception across text, vision, and audio, yet they often struggle with structured cross-modal reasoning, particularly when integrating audio and visual signals. We introduce EchoInk-R1, a reinforcement learning framework that enhances such reasoning in MLLMs. Built upon the Qwen2.5-Omni-7B foundation and optimized with Group Relative Policy Optimization (GRPO), EchoInk-R1 tackles multiple-choice question answering over synchronized audio-image pairs. To enable this, we curate AVQA-R1-6K, a dataset pairing such audio-image inputs with multiple-choice questions derived from OmniInstruct-v1. EchoInk-R1-7B achieves 85.77% accuracy on the validation set, outperforming the base model, which scores 80.53%, using only 562 reinforcement learning steps. Beyond accuracy, EchoInk-R1 demonstrates reflective reasoning by revisiting initial interpretations and refining responses when facing ambiguous multimodal inputs. These results suggest that lightweight reinforcement learning fine-tuning enhances cross-modal reasoning in MLLMs. EchoInk-R1 is the first framework to unify audio, visual, and textual modalities for general open-world reasoning via reinforcement learning. Code and data are publicly released to facilitate further research.
title EchoInk-R1: Exploring Audio-Visual Reasoning in Multimodal LLMs via Reinforcement Learning
topic Computer Vision and Pattern Recognition
Audio and Speech Processing
url https://arxiv.org/abs/2505.04623