Yuanhan Zhang*, Yunice Chew*, Yuhao Dong, Aria Leo, Bo Hu, Ziwei Liu
GPT-4o: 36.6%. Humans: 84.3%. Benchmarks video reasoning without key-frame sampling bias.
Yuanhan Zhang, Jinming Wu, Wei Li, Bo Li, Zejun Ma, Ziwei Liu, Chunyuan Li
1.2M+ HuggingFace downloads. Fully open-sourced video LMM.
Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Yanwei Li, Ziwei Liu, Chunyuan Li
One model, three scenarios — SOTA on single-image, multi-image, and video.
Yuan Liu*, Haodong Duan*, Yuanhan Zhang*, Bo Li*, Songyang Zhang*, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, Kai Chen, Dahua Lin
The go-to benchmark for evaluating multimodal models — adopted by Qwen-VL, InternVL, LLaVA, and more.
Bo Li*, Yuanhan Zhang*, Liangyu Chen, Jinghao Wan, Fanyi Pu, Jingkang Yang, Chunyuan Li, Ziwei Liu
Pioneered in-context learning for multimodal understanding. 2.8M instruction-response pairs.
Yuanhan Zhang, Kaiyang Zhou, Ziwei Liu
Pioneered in-context learning for multimodal generation. Retrieval beats random by a large margin.
Yuanhan Zhang, Kaiyang Zhou, Ziwei Liu
Where to insert LoRA, VPT, or Adapter? Let search decide.
Yuanhan Zhang, Qinghong Sun, Yichun Zhou, Zexin He, Zhenfei Yin, Kun Wang, Lv Sheng, Yu Qiao, Jing Shao, Ziwei Liu
4x ImageNet, 2x Object365. Model pre-trained on it won 1st place at ECCV 2022 CV-in-the-Wild Challenge.
Yuanhan Zhang, Zhenfei Yin, Jing Shao, Ziwei Liu
21 realms, zero overlap. No more cherry-picking benchmarks for your CLIP-like model.
