Using own proposals to test a trained faster r-cnn model

[RJVisee44]

Hi all, I've trained a faster-RCNN network on my own dataset using the RPN. I just want to test (to test the proposal generation) my own proposals on a few images. It seems that you should just be able to set cfg.TEST.HAS_RPN = False and then use scores, boxes = im_detect(net, im, obj_proposals) (to test on a single image), but then i run into an issue where:

raise Exception('Input blob arguments do not match net inputs.')
       Exception: Input blob arguments do not match net inputs.

I've also tried using the fast r-cnn "test.prototxt", but then obviously the trained network and the fast r-cnn network don't match up. I'm assuming I have to retrain without the RPN, but is there a way to get around this issue without having to retrain? I feel like this should be a simple test.

[alice-cool]

Do you solve the problem, I also want to compliment the remand of the network structure. If you have methods, please help me.