On training my own data set in CellPose

I’m trying to train CellPose with my own images, but it doesn’t seem to be working as follows.

Epoch 7, Time 1682.5s, Loss nan, Loss Test nan, LR 0.1556

The data are in the format of data science bowl 2018 and the images are grayscale.
I am working on jupyter notebook.

train_images = []
train_mask_images = []
for i in tqdm(range(len(train_image_paths))):
    train_images.append(tifffile.imread(train_image_paths[i]))
    train_mask_images.append(tifffile.imread(train_mask_paths[i]))
test_images = []
test_mask_images = []
for i in tqdm(range(len(test_image_paths))):
    test_images.append(tifffile.imread(test_mask_paths[i]))
    test_mask_images.append(tifffile.imread(test_mask_paths[i]))
model = models.CellposeModel(gpu=True, model_type='nuclei')
channels=[0,0]
model.train(train_data=train_images, train_labels=train_mask_images, train_files=train_image_paths, test_data=test_images, test_labels=test_mask_images, test_files=test_image_paths, channels=channels, save_path='data/train/models'