Minor doc issue: clarify if all images need to be the same size

[dgtlmoon]

https://www.deepdetect.com/platform/docs/object-detection/

I don't feel like it states clearly if

• Images of multiple sizing are allowed in one training set (I have anything 500x200 to 1200x768 etc)
• If it will resize them all down to the width and height settings INCLUDING the bbox training square

[dgtlmoon]

resize-all-images-to-largest-image-in-dir.py.txt

[dgtlmoon]

This is important to know, so that we know if we can use loads of existing datasets out there like https://www.kaggle.com/andrewmvd/dog-and-cat-detection which have different sizing

[dgtlmoon]

from beniz "you can define the size at training time with width and height to use larger images than 300x300. The size is then fixed yes."

yeah this is not super obvious on that page

• If training images are less than the network size, and they are a mix of resolutions, you dont need todo anything.
• if training images are greater than the network size, they all need to be the same size because you will need to use width and height which will scale the image and the bbox down to the network size