where are the tif format labels?

[selimsef]

where are the tif format labels?

[selimsef]

Labels were generated from original masks. They just represent labeled ground truth masks as a single image, i.e. single channel, different object have different pixel values. PNG could be used as well provided that there are no more than 255 objects on a single image.

[LXYTSOS]

And could you make a list of how many arguments should I pass to train.py and its meaning? It seems a lot of arguments that used for training are not passed to train.py according to train_all.sh such as alias, schedule, etc.

[selimsef]

Default parameters are specified in the file params.py

[LXYTSOS]

Thank you very much.

[LXYTSOS]

I don't quite understand these codes in dsb_binary.py:
img_id = all_ids[i] msk = cv2.imread(os.path.join(masks_dir, '{0}.png'.format(img_id)), cv2.IMREAD_UNCHANGED) lbl = cv2.imread(os.path.join(labels_dir, '{0}.tif'.format(img_id)), cv2.IMREAD_UNCHANGED) tmp = np.zeros_like(msk[..., 0], dtype='uint8') tmp[1:-1, 1:-1] = msk[1:-1, 1:-1, 0] good4copy = list(set(np.unique(lbl[lbl > 0])).symmetric_difference(np.unique(lbl[(lbl > 0) & (tmp == 0)])))
what are the img_id.png and img_id.tif here? and why were you do the symmetric_difference?

[selimsef]

This code is related to nuclei copy augmentation

[LXYTSOS]

Could you be more specific about how to do the nuclei augmentation and create those img_id.png masks and img_id.tif labels?