Explorations on facial image keypoint detection.

This project contains the explorations on face feature detection (for the Kaggle competition), as part of the final project for the 207-Machine Learning course for Berkeley's Master in Information and Data Science.

The project was developed by: Alex, Ankit, Nina and Will.

We created a Prezi adventure to showcase our ideas and explorations.

The notebook FacialKeypointDetection-AlexAnkitNinaGuillermo located in scripts/final-notebook/ contains the final report summarizing most of the work included in the rest of the repository.

For a full exposition refer to this notebook. However, this notebook is better treated as a report, than a runnable notebook, due to its size and its lengthy write-ups. For details on the runned codes refer to the specific notebooks, as guided by the READMEs.

• scripts/

  • explorations/ contains dataset explorations, plots and initial modeling attempts

  • preprocessors/ contains the preprocessing scripts

  • modelers/ contains all serious modeling attempts, including the generation of their submission files

  • tools/ contains a set of additional tools that make the exploration and modeling cleaner. For example submit.py contains functions for creating the submission files in the appropriate folder

• data/

  • datasets/ contains the original kaggle data

  • submissions/ contains the csv files submitted to the Kaggle competition

  • models/ contains the persistent storages of the models created. Each pickled model contains: name, alias, description, model-object, prediction-df, [training-time], [predicting-time]

  • preprocessed/ contains preprocessed datasets. For temporal time-consuming preprocessings

• First of all clone the repo's folder:

$ git clone https://github.com/WillahScott/facial-keypoint-detection.git

Use of a virtual environment is highly recommended (specially through conda).

Should you choose to not create a virtualenv and just install directly on your raw machine just install the prereqs (follow step 2 for virtualenv instructions)

• Clone the environment as provided in environment.yml:

$ conda env create -f environment.yml
$ source activate fkd

That's it!
For more info on using virtual environments with conda see here

• Create a virtual env (from within the folder) and activate it:

$ cd facial-keypoint-detection
$ virtualenv fkd
$ source fkd/bin/activate

• Install pre-reqs:

$ pip install -r requirements.txt

Refer to the subfolders README's for more details on the sections, contents and usage.

Last update: April, 2016