Closed RyanHuangNLP closed 5 months ago
The training data is from ShareGPT4V, you could refer to this link: https://sharegpt4v.github.io/
We will release the training code soon
@beichenzbc I'm Looking forward to the open-sourcing of the training code.
By the way, will the urban-200 evaluation set be open-sourced? This long caption evaluation dataset is significant. Thank you again for this work.
Thanks for recoginition. The training code has been released. We will open-source our urban-200 evaluation set, but it may take some time.
1.Is the training data could generate by mllm, get the dense caption? 2.how to select the training data, simple image is hard to generate long caption