Questions about ct-clip
I would like to ask if ct-clip is based on openai's clip big framework, or did you create a new one yourself?
Hi @QuiNiang, it is similar to OpenAI's framework, but the image encoder and text encoder have been modified. In the vanilla CLIP model, ViT can only accept 2D images, whereas we work with 3D volumes. Additionally, we use a pretrained language model that was trained on radiology reports. The training framework remains the same.
Hi，@sezginer ，thank you for your reply, it gives me a better understanding of the CT-CLIP model, have a nice life.
Hi，@sezginer ，the pretrained models link provided on github is not opening. Clicking on the link leads to the message Entry not found. how do I get this file?
Thanks.
Hi @QiuNiang , I fixed the links in GitHub. Also please see the folder in this repository for the models: https://huggingface.co/datasets/ibrahimhamamci/CT-RATE/tree/main/models. Models deprecated is for the v1.