search

zhaoziheng / SAT

The official repository for "One Model to Rule them All: Towards Universal Segmentation for Medical Images with Text Prompts"
145 stars 7 forks source link

Will Training code be resleased？ #1

Closed sharonlee12 closed 9 months ago

sharonlee12 commented 9 months ago

Hello! It's really an excellent work! May I ask if there will be a training code release in the future？Also，how do you exploit the domain knowledge, namely, e-Anatomy2 and UMLS?

sharonlee12 commented 9 months ago

besides,May I ask where I can get pubmed_bert,for: model = Text_Tower('../medical-knowledge-pretraining/src/others/pubmed_bert', 768)

zhaoziheng commented 9 months ago
  1. You can download the tokenizer by adding this to the inference command: --tokenizer_path 'microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract'.
  2. Yes, we will release the training code in the future.
  3. You can refer to our paper for details of domain knowledge encoding.
sharonlee12 commented 9 months ago

  1. --tokenizer_path 'microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract'

  2. You can download the tokenizer by adding this to the inference command: --tokenizer_path 'microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract'.

  3. Yes, we will release the training code in the future.

  4. You can refer to our paper for details of domain knowledge encoding.

Thank you very much！