Faster RCNN PyTorch Download, Train and Test on COCO 2014 dataset

1) Get the files from Ruotian Luo's github repository.
     git clone https://github.com/ruotianluo/pytorch-faster-rcnn.git
2) Get into the faster rcnn directory
     cd pytorch-faster-rcnn/
3) Determine your achitecture. My GPU model is nVidia Tesla P100 and  so the corresponding architecture according to this website is sm_60. The architecture will be provided to set the Nvidia Cuda Compiler's(nvcc) -arch flag in the 4th step below.
4) Compile and Build RoI Pooling module
     cd lib/layer_utils/roi_pooling/src/cuda
     nvcc -c -o roi_pooling_kernel.cu.o roi_pooling_kernel.cu -x cu -Xcompiler -fPIC -arch=sm_60
     cd ../../
     python build.py
     cd ../../../
5) Compile and Build Non-maximum Suppression(NMS) module
     cd lib/nms/src/cuda
     nvcc -c -o nms_kernel.cu.o nms_kernel.cu -x cu -Xcompiler -fPIC -arch=sm_60
     cd ../../
     python build.py
     cd ../../
 6) Install python COCO API under data folder.
     cd data
     git clone https://github.com/pdollar/coco.git
     cd coco/PythonAPI
     make
     cd ../../..
7) (OPTIONAL) In case your COCO data set(images and annotation) have to reside under a different directory you can create a symbolic link to it under data folder.
#move all files under coco (obtained after 6th step above) to shared coco folder
     mv data/coco/* /YOURSHAREDDATASETS/coco
#create symbolic link to that coco folder
     cd data
     rm -rf coco
     ln -s /YOURSHAREDDATASETS/coco coco
8) Download proposals and annotation json files from here. 
9) After you downloaded annotations, place them under coco/annotations folder. The coco folder structure should look like below. To download default COCO images and annotations please check here.

coco/
        annotations/
                             instances_minival2014.json
                             instances_valminusminival2014.json
                             ...(default coco json files)
        images/
                    train2014/...
                    val2014/...
                    test2014/...
        PythonAPI/...
10) Create imagenet_weights folder under data. A pre-trained model on ImageNet dataset will reside here.
     mkdir -p data/imagenet_weights
11) Download the pre-trained ResNet model(the resnet101-caffe one) model from here into the imagenet_weights folder and rename it.
    cd data/imagenet_weights
    mv resnet101-caffe.pth res101.pth
     cd ../..   
12)Run training script
     GPU_ID=0
     DATASET=coco
     MODEL=res101 
     ./experiments/scripts/train_faster_rcnn.sh $GPU_ID $DATASET $MODEL
13) Test your model upon successful trainig.
#change the iteration number in test_faster_rcnn.sh to
#the number of iterations you made. I made 460000 iterations the default is 490000. 
     GPU_ID=0
     DATASET=coco
     MODEL=res101 
     ./experiments/scripts/test_faster_rcnn.sh $GPU_ID $DATASET $MODEL