Beginning with VisualGLM and CogVLM, we are continuously exploring VLMs in pursuit of enhanced vision-language fusion, efficient higher-resolution architecture, and broader modalities and applications. Here we propose the CogVLM2 family, a new generation of visual language models for image and video understanding including CogVLM2, CogVLM2-Video and GLM-4V. As an image understanding model, CogVLM2 inherits the visual expert architecture with improved training recipes in both pre-training and post-training stages, supporting input resolution up to 1344×1344 pixels. As a video understanding model, CogVLM2-Video integrates multi-frame input with timestamps and proposes automated temporal grounding data construction. Notably, CogVLM2 family has achieved state-of-the-art results on benchmarks like MMBench, MM-Vet, TextVQA, MVBench and VCGBench. All models are open-sourced in https://github.com/THUDM/CogVLM2 and https://github.com/THUDM/GLM-4, contributing to the advancement of the field.
@article{arxiv.2408.16500,
title = {CogVLM2: Visual Language Models for Image and Video Understanding},
author = {Wenyi Hong and Weihan Wang and Ming Ding and Wenmeng Yu and Qingsong Lv and Yan Wang and Yean Cheng and Shiyu Huang and Junhui Ji and Zhao Xue and Lei Zhao and Zhuoyi Yang and Xiaotao Gu and Xiaohan Zhang and Guanyu Feng and Da Yin and Zihan Wang and Ji Qi and Xixuan Song and Peng Zhang and Debing Liu and Bin Xu and Juanzi Li and Yuxiao Dong and Jie Tang},
journal= {arXiv preprint arXiv:2408.16500},
year = {2024}
}