RadioFormer Inference on LLD-MMRI dataset

Data and Model

• Download the proprocessed croped tumor volumes at (Baidu Disk pw:299z)
• Download the model at (Baidu Disk pw: kmfb)

install requirement

pip install -r requirements.txt

Validation code

run

python validation_new.py
 --data_dir
/root_to_image_path/images_mycrop_8_challenge
--val_anno_file
/root_to_label_path/labels/labels-validation.txt
--model
radioformer_vit_small
--batch-size
1
--checkpoint
/root_to_ckpt_path/ckpts/fold1.pth.tar
--results-dir
/path_to_save/results
--score-dir
/path_to_save/results

results

Using the given model, you will obtain results on the LLD-MMRI Phase I validation set as follows:

on fold 1:

on fold 2:

on fold 3:

on fold 4:

on fold 5:

which match the results reported in Table 1 of our MICCAI paper.