Leveraging medical Twitter to build a visual–language foundation model for pathology AI

This Github repository contains the evaluation codes.

Links

• Official huggingface website
• PLIP model (huggingface transformers)
• Preprint

Config Env File

PC_CACHE_FOLDER: is the folder in which cached embeddings are saved
PC_RESULTS_FOLDER: is the folder in which results are going to be saved
PC_EVALUATION_DATA_ROOT_FOLDER: this folder should point to the folder in which there are the evaluation dataset
PC_DEFAULT_BACKBONE: this should point at the backbone to use as default for PLIP
PC_CLIP_ARCH: this is the architecture for PLIP and BLIP (e.g., "ViT-B/32")

Evaluation Scripts

All evaluation scripts are found in path_eval/scripts. They use the abstractions found in path_eval/evaluation

Dataset Details

Data Location (For internal use only)

Classification Task /v2/evaluation_datasets/classification

Retrieval Task /v2/evaluation_datasets/image_retrieval

Setup Images

Data has been generated by splitting using: seed=1_trainratio=0.70_size=224