/Faster-RCNN-TensorFlow

TensorFlow implementation of Faster RCNN for Object Detection

Primary LanguagePythonMIT LicenseMIT

Faster-RCNN-TensorFlow

Language python TensorFlow License

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
  1. 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.

GPU model Architecture
TitanX (Maxwell/Pascal) sm_52
GTX 960M sm_50
GTX 1080 (Ti) sm_61
Grid K520 (AWS g2.2xlarge) sm_30
Tesla K80 (AWS p2.xlarge) sm_37

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

Classes AP
aeroplane 0.7391
bicycle 0.7803
bird 0.6681
boat 0.5576
bottle 0.5236
bus 0.7661
car 0.8000
cat 0.7840
chair 0.4995
cow 0.7252
diningtable 0.6721
dog 0.7504
horse 0.7843
motorbike 0.7410
person 0.7739
pottedplant 0.4401
sheep 0.6616
sofa 0.6519
train 0.7431
tvmonitor 0.7106
mAP 0.6886

Release: v0.5.1