Update demo.py

Fix to use different resnet architectures based on input

Current code always uses resnet 101 architecture for all resnet based architectures while ignoring the arguments for current architecture

Update bbox_transform.py

Fix typo in faster_rcnn.py

Fix typo in faster_rcnn.py

Remove "for training" in get_roidb

get_roidb will be indirectly called by test_net.py.
This PR removes "for training" because it's misleading.

Fix issue #452 for PyTorch 0.4.0

two pdb imports is redundant

you import pdb twice.. also restructured the whitespace to reflect trainval_net.py

Fix rgb -> bgr while image loading and displaying

cap.read() returns BGR image.
So rgb->bgr is only needed for scipy.misc.imread 

cv2.imshow expects BGR image for displaying properly(RGB)
Refer: http://stackoverflow.com/a/15074748/1134940

typo fixes

typo fix

Typo Fix

Fix to cache image sizes data

fix bug

fix test_net.py run into exceptions when __C.TEST.BBOX_REG = False

fix run into exceptions when __C.TEST.BBOX_REG = False

fixed typo

typo fixed

typo fix

fixes user warning in demo.py