Real-world Image Recognition for Multiple Human Attributes

This project is participating in NUS SoC 13th STePS. For more information, see here.

Information about the repository structure:

• Project Report
• Gender Classifier for the LFW dataset
  • Implemented using PyTorch (https://pytorch.org) library.
  • An older model can be found at here.
• Long/short-sleeves Classifier for the WIDER dataset
  • Implemented using TensorFlow (https://www.tensorflow.org) library.

Notes regarding the WIDER dataset

See here.

Notes regarding the LFW dataset

See here.

Using Docker to install Tensorflow

• Sometimes, it can be quite troublesome to install tensorflow on your machine. If so, you can use Docker to release the pain.
• Pull the Docker image by docker pull tensorflow/tensorflow:latest-py3
• Let's first create a container by docker run -it --name tensorflow -p 0.0.0.0:7007:6006 -v $PWD:/CS3244 tensorflow/tensorflow:latest-py3 bash.
  • it is always used for interactive command, like a shell.
  • rm will delete the container automatically after exit.
  • v and w specifies how the directory sharing is done.
  • p will do port forwarding (for tensorboard later). Use tensorboard --logdir ./ --host 0.0.0.0 to run tensorboard. Then you can see the board at http://localhost:7007.
• From then on, you can start the container by docker start tensorflow. Then, you can login by docker exec -it tensorflow bash. To stop the container, run docker stop tensorflow.

Resources

• http://mmlab.ie.cuhk.edu.hk/projects/WIDERAttribute.html
• https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/
• http://vis-www.cs.umass.edu/lfw/

Licence

GNU General Public Licence 3.0