How to deal with dataset with different size image

As the question,I’m building a CNN, I got a dataset with different size images, for example size=198 * 256, size = 210 * 220, etc. I want use tt.RandomCrop to improve my model, but I’m confused what size I should take in tt.RandomCrop, should I zoom those picture to a fixed size or do something else. I’m just starting learning deep learning, any kind of hint is welcome.

Well, it is almost always the case when you want to resize your images to a concrete shape,
for example 224x224 (worth mentioning, that it is highly depends on which size your test images have).

I’ve used resizing too, when I encountered datasets with different image shapes.

By doing so, now you can clearly understand, which shape an input has and build your network.

For resizing you can you image = cv2.resize(image, new_size), see also
OpenCV resize function documentation