Running out of GPU memmory

[kadattack]

Is there a way to decrease image size through the config file to lower required memory?
Decreasing batch size IMS_PER_BATCH helps but it doesn't seem to work with MIN_SIZE_TRAIN and MAX_SIZE_TRAIN variables.
Is there a variable you can set so that it will transform the images down to lower resolution before loading in to GPU?

[kadattack]

I really am baffled with how this is working. I manually reduced the sizes of the image and the masks and yet still the RAM consumption stayed the same?? It took the same amount of memory whether the images where 512x384 or 288x216 or whether i reduced the MIN_SIZE_TRAIN and MAX_SIZE_TRAIN variables.
If none of this works, what exactly is affecting the size of the images being trained on your GPU?