Does Google AutoML Vision normalize images?

Question

I would expect normalization to be done according to best practices, but there's no documentation on that.

According to this answer, AutoML does augmentation behind the scenes, but there's no documentation on that.

Answer 1

As per the official documentation the suggested maximum is 1024 pixels by 1024 pixels. Images much larger than 1024 pixels by 1024 pixels there may be some image quality loss during AutoML Vision Object Detection's image normalization process. Please review Preparing your training data