Input-Images pre-processing confusion in keras-retinanet

[hlmhlr]

Hi,

I have few basic confusions regarding input data processing in keras-retinanet, just want to get clarify regarding the following:

1. As Resnet-50 network requires smaller size of input (normally 256x256), and the min_max size in resnet-retinanet is 1388x800. So, how these sizes are matched in the pre-processing section?

2. How can we visualize charaeteristics (shape, size, resolution) of processed input images before training.

3. How can we determine the aspect ratio in the case we have different sizes of input images.

4. Also, apart from image_preprocess() and resize_image() function, which additional functions have been used in the case we do not define any input size of image? and, specially, how i can see these outputs before the start of training?

5. For my dataset, i have images with different resolution and sizes, and also every image contains multiple small objects, So i am confused how to process them rightly for the smooth training. Therefore, have written above queries.

I am sorry for basics since I am beginner in this, i would be grateful for positive response.

Thanks,