producing image embeddings using own image dataset

[p1nkuu]

Hello, is there a way to produce image embeddings from BiomedGPT directly or should I use the OFA or CLIP model directly and just use your checkpoints to train it? I would like to create image embeddings for my own dataset of histopathology images. Sorry if this is a silly question, I'm new to using large models like this!

[taokz]

@p1nkuu Sorry for the late response. I currently don’t have the code to retrieve image embeddings, but I plan to develop it in the following week. Recently, there was a paper that used BiomedGPT as one of the biomedical embedding models: https://arxiv.org/abs/2410.06542. You might find it useful as a reference.