I have been building a multilabel image classification model using inception v3, which uses images of size 299x299, I have been wondering what are the effects of feeding images of rectangular shapes for example (or arbitrary resolutions) on the performance of the model, and if I can define requirements for how the data should be to ensure optimal performance, what would those requirements be ? Intuitively, I think that square images would perform better than rectangular images, is this true?



You must log in to answer this question.

Browse other questions tagged .