Faster-RCNN-TensorFlow

This is an experimental TensorFlow implementation of Faster-RCNN, based on the work of smallcorgi and rbgirshick. I have converted the code to python3, future python2 will stop supporting it, and using python3 is an irreversible trend. And I deleted some useless files and legacy caffe code.

What's New:

• Convert code to Python3
• Make compile script adapt gcc-5
• Visualization using tensorboard
• PSRoI Pooling
• OHEM a.k.a Online Hard Example Miniing
• RoI Align
• ResNet50
• PVANet
• MobileNet v1

Reference:

Acknowledgments:

1. py-faster-rcnn

2. Faster-RCNN_TF

3. ROI pooling

4. TFFRCNN

Installation (sufficient for the demo)

1. Clone the Faster R-CNN repository

git clone https://github.com/walsvid/Faster-RCNN-TensorFlow.git

2. Build the Cython modules

   ROOT = Faster-RCNN-TensorFlow
   cd ${ROOT}/lib
   make

Compile cython and roi_pooling_op, you may need to modify make.sh for your platform.

Demo

After successfully completing basic installation, you'll be ready to run the demo.

Download model training on PASCAL VOC 2007 [Google Drive]

To run the demo execute:

cd $ROOT
python ./tools/demo.py --model model_path

The demo performs detection using a VGG16 network trained for detection on PASCAL VOC 2007.

Train model

1. Download the training, validation, test data and VOCdevkit

   wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
   wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
   wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCdevkit_08-Jun-2007.tar

2. Extract all of these tars into one directory named VOCdevkit

   tar xvf VOCtrainval_06-Nov-2007.tar
   tar xvf VOCtest_06-Nov-2007.tar
   tar xvf VOCdevkit_08-Jun-2007.tar

3. It should have this basic structure

   $VOCdevkit/                           # development kit
   $VOCdevkit/VOCcode/                   # VOC utility code
   $VOCdevkit/VOC2007                    # image sets, annotations, etc.
   # ... and several other directories ...

4. Create symlinks for the PASCAL VOC dataset

   cd $FRCN_ROOT/data
   ln -s $VOCdevkit VOCdevkit2007

5. Download pre-trained ImageNet models

   Download the pre-trained ImageNet models [Google Drive]

   mv VGG_imagenet.npy $FRCN_ROOT/data/pretrain_model/VGG_imagenet.npy

6. Run script to train and test model

   cd $FRCN_ROOT
   ./experiments/scripts/faster_rcnn_end2end.sh $DEVICE $DEVICE_ID VGG16 pascal_voc

   DEVICE is cpu or gpu. Please note that if CUDA_VISIBLE_DEVICES is used as the mask for the specified GPU, please note the GPU ID. If it is an example like this: CUDA_VISIBLE_DEVICES=1 ./experiments/scripts/faster_rcnn_end2end.sh gpu 1 VGG16 pascal_voc, which means that the GPU1 is used as the starting number 1+1 or GPU2. If there are not multiple GPUs, use GPU0. Another example: use CUDA_VISIBLE_DEVICES=1 ./experiments/scripts/faster_rcnn_end2end.sh gpu 0 VGG16 pascal_voc so that although the displayed gpu id is 0, actually used is 1+0=GPU1.

Visualization

Just execute tensorboard.

tensorboard --logdir=./logs

The result of testing on PASCAL VOC 2007

VGG16

Release: v0.5.1