How do I change my channel to 1 to have a better performence on trainig with grayscale datasets?

[subwheel]

I got a dataset of grayscale images to train.
At first, I use the model's default setting (3 channels for RGB, I guess), and the result couldn't satisfy me.
Then I do some research, finding that changing my channels to only 1 might lead to a better performence on grayscale images.
However, I could not find a place to alter my channels, neither any tutorials.
Anyone knows how change the number of channels?
Any tips for grayscale images training are welcomed, too.
Thanks.