rcgalbo / TF_ObjectDetection_API

Tutorial on how to create your own object detection dataset and train using TensorFlow's API

TensorFlow Object Detection API Tutorial

This repository has the code from my O'Reilly article published on October 25, 2017.

Required Packages

• TensorFlow v1.3
• Jupyter
• NumPy
• Scipy
• Matplotlib
• Scikit-Image
• Pandas
• lxml
• protobuf

There are two ways you can install these packages: by using Docker or by using native Python 3.5.

Using Docker

1. Download and install Docker. If using Ubuntu 14.04/16.04 I wrote my own instructions for installing docker here.

2. Download and unzip this entire repo from GitHub, either interactively, or by entering

   git clone https://github.com/wagonhelm/TF_ObjectDetection_API.git
   

3. Open your terminal and use cd to navigate into the directory of the repo on your machine

   cd TF_ObjectDetection_API

4. To build the Dockerfile, enter

   docker build -t object_dockerfile -f dockerfile .

   If you get a permissions error on running this command, you may need to run it with sudo:

   sudo docker build -t object_dockerfile -f dockerfile .

5. Run Docker from the Dockerfile you've just built

   docker run -it -p 8888:8888 -p 6006:6006 object_dockerfile bash

   or

   sudo docker run -it -p 8888:8888 -p 6006:6006 object_dockerfile bash

   if you run into permission problems.

6. Install TensorFlow Object Detection API

   cd models/research/
   protoc object_detection/protos/*.proto --python_out=.
   export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
   cd ..
   cd ..

7. Launch Jupyter and Tensorboard both by using tmux

   tmux
   
   jupyter notebook --allow-root

   Press CTL+B then C to open a new tmux window, then

   tensorboard --logdir='data'

   To switch windows Press CTL+B then window #

   Once both jupyter and tensorboard are running, using your browser, navigate to the URLs shown in the terminal output if those don't work try http://localhost:8888/ for Jupyter Notebook and http://localhost:6006/ for Tensorboard. I had issues with using TensorBoard with Firefox when launched from Docker.

Using Native Python 3

1. Install system requirements

sudo apt-get install -y git-core wget protobuf-compiler 

2. Download and unzip this entire repo from GitHub, either interactively, or by entering

git clone https://github.com/wagonhelm/TF_ObjectDetection_API.git

3. Install Python Requirement

cd TF_ObjectDetection_API
# Requires sudo if not in a virtual environment
pip3 install -r requirements.txt
pip3 install tensorflow jupyter

4. Clone TensorFlow Models Into Repository Directory and Install Object Detection API

cd TF_ObjectDetection_API
git clone https://github.com/tensorflow/models.git

You will have to run this command every time you close your terminal unless you add the the path to slim to your .bashrc file

cd models/research/
protoc object_detection/protos/*.proto --python_out=.
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
cd ..
cd ..

4. Launch Jupyter

jupyter notebook

5. Launch Tensorboard In New Terminal

tensorboard --logdir='data'

Once both jupyter and tensorboard are running, using your browser, navigate to the URLs shown in the terminal output if those don't work try http://localhost:8888/ for Jupyter Notebook and http://localhost:6006/ for Tensorboard.