As a nonprofit platform, our human and computational resources are limited. All evaluation code has already been open-sourced. If you are an algorithm developer working on PFMs and have the ability to debug and deploy models yourself, we strongly encourage you to run experiments on your own computing equipment. This helps reserve our limited resources for clinical researchers who may not have coding or debugging experience. Thank you for your support.

1. How should I prepare a task?

   You can submit a task via the “New Task” option in the left sidebar. For widely used PFMs such as UNI, Prov-GigaPath, and Virchow, simply select them from the dropdown menu. For other PFMs or your own trained models, you may either provide a link to the corresponding code repository or upload the packaged model files directly.

2. What should be included in a custom PFM?

   It should include the model definition code, pretrained weights, image preprocessing code, and a feature-extraction function or API (taking an RGB image as input and returning a 1×n feature embedding). You should also provide all required environment dependencies, preferably consolidated in a requirements.txt file. We strongly recommend following the structure used in UNI when organizing your PFM. Please note that only PyTorch-based models are supported at the moment.

3. How to contact us?

   If you encounter any issues during use—such as not knowing how to operate the platform, abnormal evaluation results, or website errors—please contact us at wplin@stu.xmu.edu.cn