Question about training parameters

[yh675]

Hi, thanks for this great work!

I am trying to train this network on my own labeled dataset. I augmented it with main.py in the MBIPED repo.

For image configuration in dataset_manager, I am confused as to what the image_width and image_height are referring to.

Are these referring to the actual sizes of the augmented images? Or do they refer to what sizes the network will downsample the image to? Or the original sizes of the images pre-augmentation?

Also for testing should I keep the same image_width and image_height as testing?

Thanks

[weinixuehao]

I also have the same quetion