madhuatberkeley / w207_project

w207 Project: Facial Keypoints Detection

Authors: Madhu Hegde and Jane Hung
Date: 6 Dec 2020

Objective

Detect facial keypoints (x-, y-coordinates) according to images of faces. Taken from this Kaggle competition.

Goal

Reduce our RMSE below our initial baseline NN model.

Folder Structure

• /input: contains training and test data
• /output: contains stored model and history data
• /reference: contains information on how to submit to Kaggle
• /source: contains model building code
• /requirements.txt: contains package requirements to recreate our findings

Reproducibility

1. Run source/w207_eda.ipynb to understand how we've explored this dataset.
2. Run source/w207_project.ipynb to run through our wip feature engineering and modeling pipeline.
3. Run source/w207_project.py to run through our finalized feature engineering and modeling pipeline.
4. Run source/w207_project_capsule.ipynb to run through the our experimental capsule network code.