Working on large sized images

Hello, I am currently doing the MOOC(lesson 2) and I was trying to apply the concepts learned in class for the competetion:Whale-categorization. One of points i noticed is that image size are larger in size like [1000,500]. So we cannot use resnet architecture here.
hist-plots
Blue: Column size
Green: Row size
This is the histogram plot i obtained from training set images according to their sizes.
What are other architectures which we can work on and also can some one suggest how to work with large images?

Most modern architectures can work with images of any size. However, it’s typical to resize the images first, for example to 256x256. If the original image is not square, you can choose to resize the shortest size to 256 (and the longest side proportionally), then crop out the center square (or a random square).

Hi @machinethink, when we provide the size in ImageClassifierData does it resize the images according to the size provided

My understanding is that it does indeed.

I don’t think ImageClassifierData itself cares about the image size, but usually it’s given a transform that does the resizing.