Transfer learning for images of single channels

vision
1

I have dataset of 75000 images of dimension1000*65 .

and another dataset of 75000 images of size 65*90.

How to train these images using transfer learning like alexnet,vgg or resent.

or I have to design and train new model for these dataset.

2

Most pre-trained models accept three channels input. The simplest way to handle this is by repeating the single channel 3 times; by, for example, applying the following transform while loading your data:

image_transfrom = transforms.Compose([
           ...
           transforms.Lambda(lambda x: x.repeat(3, 1, 1)),
            ...
            ])
3

There is also transform.Grayscale(3) that performs a grayscale tranform into 3 channels (R=G=B) images. It should work on grayscale input image.