This repo is an implementation of faster r-cnn using Keras and Tensorflow. The Tensorflow part borrows the ref: https://github.com/endernewton/tf-faster-rcnn a lot.

The implementation is aiming to build the Keras interface based on a Tensorflow FRCNN code [1]. Most of the tf functions are packed by using Keras Lambda. And the structure of the code is changed.

• Pascal VOC 2007

• Pascal VOC 2007 + 2012

• COCO 2014

• Ubuntu 16.04

• Python 2/3, in case you need the sufficient scientific computing packages, we recommend you to install anaconda.

• Tensorflow >= 1.5.0

• Keras >= 2.2.0

• Optional: if you need GPUs acceleration, please install CUDA that the version requires >= 9.0

• Data preparation and setup

  Follow the ref: https://github.com/rbgirshick/py-faster-rcnn#beyond-the-demo-installation-for-training-and-testing-models. Download Pascal VOC 07 dataset and build soft link in folder ./data/, then name the link as "VOCdevkit2007"

  To compile the lib, move into the lib folder "cd ./lib". According to your hardware, change the "-arch" parameter in setup.py line: 130. Then run

  make
  

  Tips: to find your hardware compiling setting, you can refer to: http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/. For me using the Titan XP, I set the "-arch" as "sm_61".

• Training

  Build the model weight saving folder "../output/[NET]/"

  Download pre-trained models and weights:

  mkdir net_weights
  wget https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels.h5
  cd ..
  
  

  Run the following script:

  ./scripts/train_faster_rcnn.sh [GPU_ID] [DATASET] [NET]
  # GPU_ID is the GPU you want to test on
  # NET in {vgg16} is the network arch to use,
  # DATASET {pascal_voc, pascal_voc_0712, coco} is defined in train_faster_rcnn.sh
  # Examples:
  ./scripts/train_faster_rcnn.sh 0 pascal_voc vgg16
  

• Testing

  Build the test output saving folder "../output/[NET]/"

  Run the following script:

  ./scripts/test_faster_rcnn.sh [GPU_ID] [DATASET] [NET]
  # GPU_ID is the GPU you want to test on
  # NET in {vgg16} is the network arch to use,
  # DATASET {pascal_voc, pascal_voc_0712, coco} is defined in train_faster_rcnn.sh
  # Examples:
  ./scripts/test_faster_rcnn.sh 0 pascal_voc vgg16
  
  

• [1] tf-faster-rcnn: https://github.com/endernewton/tf-faster-rcnn