CS194-26: Image Manipulation and Computational Photography

Project 5: Facial Keypoint Detection with Neural Networks

by Christy Koh, November 2021

Overview

This project aims to use neural networks to automatically detect facial keypoints.

PART I: NOSE TIP DETECTION

Custom Dataloader

This first part uses the IMM Face Database to train a toy model for nose tip detection. Pictured below are three sampled images from the dataloader, visualized with ground-truth nose tip keypoints.

Convolutional Neural Network

For this model, I used 3 convolutional layers and 2 fully-connected layers, with ReLu and max-pooling after each convolutional layer.

The loss function was a simple Mean-Squared Error (nn.MSELoss) computation, and we used the Adam gradient-based optimizer to update the