How to apply train.py for an one-channel dataset?

[LeeBC2298]

I tried to use cv2.IMREAD_COLOR to train a three-channel-input model for FMNIST dataset, and this method works well.
However, when I tried to use cv2.IMREAD_GRAY to train a one-channel-input model for FMNIST dataset, this method dosen't work.
I have already changed 'nc' from 3 to 1 for the generator, and changed 'true_grad.view(-1, 3072)' to 'true_grad.view(-1, 1024)' for the one channel model setting. What else should I do?