ParGo: Bridging Vision-Language with Partial and Global Views

This work presents ParGo, a novel Partial-Global projector designed to connect the vision and language modalities for Multimodal Large Language Models (MLLMs). Unlike previous works that rely on global attention-based projectors, our ParGo bridges the representation gap between the separately pre-trained vision encoders and the LLMs by integrating global and partial views, which alleviates the overemphasis on prominent regions. To facilitate the effective training of ParGo, we collect a large-scale detail-captioned image-text dataset named ParGoCap-1M-PT, consisting of 1 million images paired with high-quality captions. Extensive experiments on several MLLM benchmarks demonstrate the effectiveness of our ParGo, highlighting its superiority in aligning vision and language modalities. Compared to conventional Q-Former projector, our ParGo achieves an improvement of 259.96 in MME benchmark. Furthermore, our experiments reveal that ParGo significantly outperforms other projectors, particularly in tasks that emphasize detail perception ability.

Authors

• An-Lan Wang SUN YAT-SEN UNIVERSITY ByteDance China
• Bin Shan ByteDance China
• Wei Shi Rocket Force University of Engineering
• Kun-Yu Lin SUN YAT-SEN UNIVERSITY
• Xiang Fei ByteDance China
• Guozhi Tang ByteDance China
• Lei Liao ByteDance China
• Jingqun Tang ByteDance Inc., Hangzhou, China
• Can Huang ByteDance China
• Wei-Shi Zheng SUN YAT-SEN UNIVERSITY Key Laboratory of Machine Intelligence and Advanced Computing, Ministry of Education, China

Keywords

No keywords are indexed for this paper.