Faster R-CNN code example

python main.py PATH_TO_DATASET

Things to add/change/consider

where to handle the image scaling. Need to scale the annotations, and also RPN filters the minimum size wrt the original image size, and not the scaled image
should image scaling be handled in FasterRCNN class?
properly supporting flipping
best way to handle different parameters in RPN/FRCNN for train/eval modes
uniformize Variables, they should be provided by the user and not processed by FasterRCNN/RPN classes
general code cleanup, lots of torch/numpy mixture
should I use a general config file?

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
data		data
.gitattributes		.gitattributes
.gitignore		.gitignore
README.md		README.md
faster_rcnn.py		faster_rcnn.py
gen_map.py		gen_map.py
generate_anchors.py		generate_anchors.py
mAVP_test.ipynb		mAVP_test.ipynb
main.py		main.py
model.py		model.py
pascal3d+_test.ipynb		pascal3d+_test.ipynb
pascal3d_plus.py		pascal3d_plus.py
py_cpu_nms.py		py_cpu_nms.py
roi_pooling.py		roi_pooling.py
rpn.py		rpn.py
utils.py		utils.py
voc.py		voc.py
voc_eval.py		voc_eval.py