I'm working on a project about image recognition. In my dataset I have images of different size, all rectangular image (the most 640x480 and 1280x640). I would like to build my classifier to recognize subjects of different size and shape.
Many books and references suggest to rescale all the images to be squared. Why do we need to get squared images? If I use a squared image input approach I'm afraid to get distorted view/image. I'm trying set up a CNN with rectangular input shape but I would like to hear your opinion. Sometimes people suggests to follow the "sliding window" approach, but I don't know if it's better to simply consider the rectangle input shape.
What's your experience? How do you deal with different input shapes?