The OpenCompass team of Shanghai Artificial Intelligence Laboratory Sinan and ModelScope jointly launched a major update of the multi-modal large model competition platform Compass Multi-Modal Arena! The platform aims to provide users with a convenient platform to experience and compare various mainstream multi-modal large models, and ultimately help users find the model that best meets their needs. The editor of Downcodes will introduce this exciting update to you in detail.
The OpenCompass team of Shanghai Artificial Intelligence Laboratory Sinan and ModelScope recently announced that their large model evaluation platform Compass Arena has undergone an important update and launched a new multi-modal large model competition section Compass Multi-Modal Arena. This new section provides a platform for users to experience and compare the effects of a variety of mainstream multi-modal large models, helping users find the model that best suits their needs.
The official website and ModelScope page of Compass Multi-Modal Arena have been opened to the public, providing a simple and easy-to-use interface. Users can upload images and enter questions, and the system will arrange two anonymous multi-modal large models to generate answers based on the input content. Users make subjective evaluations based on the quality of the generated content, choosing the model they believe performs better. After the evaluation is complete, the user can see the name of each model.
The platform also has a built-in special question bank, which is convenient for users to use when uploading images is inconvenient. The question bank focuses on subjective visual question and answer tasks, such as meme understanding, artwork appreciation, and photography appreciation. This design aims to evaluate the performance and user experience of multi-modal large models on subjective tasks.
Compass Multi-Modal Arena official website
https://opencompass.org.cn/arena?type=multimodal
ModelScope page:
https://modelscope.cn/studios/opencompass/CompassArena
HuggingFace page
https://huggingface.co/spaces/opencompass/CompassArena
OpenCompass multimodal evaluation tool open source link:
https://github.com/open-compass/VLMEvalKit
All in all, the update of Compass Multi-Modal Arena provides a new and convenient platform for the evaluation and selection of multi-modal large models, which is worthy of user experience and attention. We look forward to continued updates of this platform in the future to bring more surprises to users!