ask-anything

Ask-Anything is an open-source project focused on video understanding and dialogue, developed by the Shanghai Artificial Intelligence Laboratory. It provides multiple video dialogue models, including VideoChat, VideoChat2, and their variants, supporting integration with language models such as ChatGPT, StableLM, MOSS, and MiniGPT-4. The project includes features like end-to-end dialogue, long video processing, and high-resolution video understanding, and has released the MVBench benchmark and 2 million instruction data entries. The latest version, VideoChat3, shows significant improvements in general, long-video, and streaming video understanding. The project offers online demos, Hugging Face spaces, and detailed documentation, suitable for researchers and developers.