Using Our Models for Inference

To use our trained models pretrained on AbdomenAtlasMini to conduct inference on CT images, please first organize the file structures in your RESULTS_FOLDER/nnUNet/3d_fullres/ as follows. You can download the necessary files from our Baidu Netdisk Download or Google Drive link:

- Task200_AbdomenAtlasMini/
  - STUNetTrainer_small_ep2k__nnUNetPlansv2.1/
    - plans.pkl
    - fold_all/
      - model_final_checkpoint.model
      - model_final_checkpoint.model.pkl
  - STUNetTrainer_base_ep2k__nnUNetPlansv2.1/
    - plans.pkl
    - fold_all/
      - model_final_checkpoint.model
      - model_final_checkpoint.model.pkl
  - STUNetTrainer_large_ep2k__nnUNetPlansv2.1/
    - plans.pkl
    - fold_all/
      - model_final_checkpoint.model
      - model_final_checkpoint.model.pkl
  - STUNetTrainer_huge_ep2k__nnUNetPlansv2.1/
    - plans.pkl
    - fold_all/
      - model_final_checkpoint.model
      - model_final_checkpoint.model.pkl

To conduct inference, you can use following command (base model for example):

nnUNet_predict -i INPUT_PATH -o OUTPUT_PATH -t 200 -m 3d_fullres -f all -tr STUNetTrainer_base_ep2k

For much faster inference speed with minimal performance loss, it is recommended to use the following command:

nnUNet_predict -i INPUT_PATH -o OUTPUT_PATH -t 200 -m 3d_fullres -f all -tr STUNetTrainer_base_ep2k --mode fast --disable_tta

The categories corresponding to the label values can be found in the label_orders file within our repository