https://github.com/abhinav-codealchemist/facial_keypoint_detection

Facial keypoints detection using 15 landmark points on human face
https://github.com/abhinav-codealchemist/facial_keypoint_detection

computer-vision convolutional-neural-networks deep-learning jupyter-notebook keras machine-learning opencv python regression

Last synced: 3 months ago
JSON representation

Facial keypoints detection using 15 landmark points on human face

Host: GitHub
URL: https://github.com/abhinav-codealchemist/facial_keypoint_detection
Owner: abhinav-codealchemist
Created: 2019-03-03T14:50:29.000Z (about 6 years ago)
Default Branch: master
Last Pushed: 2019-04-20T17:22:52.000Z (about 6 years ago)
Last Synced: 2025-01-13T09:34:38.153Z (4 months ago)
Topics: computer-vision, convolutional-neural-networks, deep-learning, jupyter-notebook, keras, machine-learning, opencv, python, regression
Language: Jupyter Notebook
Size: 17.8 MB
Stars: 0
Watchers: 1
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

[![HitCount](http://hits.dwyl.io/abhinav-codealchemist/Facial_Keypoint_Detection.svg)](http://hits.dwyl.io/abhinav-codealchemist/Facial_Keypoint_Detection)
# Facial Keypoints Detection

I have made a Facial Keypoints Regressor that regresses the keypoints for human face. This keypoint detection problem is a regression problem since we are predicting the coordinates for the landmarks on the face which are real numbers.
Detection of facial keypoints is very useful for a number of tasks like facial recognition, detection etc. The keypoints selected in this code uses 15 keypoints representing the various coordinates on the human face.
The dataset for this problem was downloaded from Kaggle https://www.kaggle.com/c/facial-keypoints-detection/.

CNN was used for this problem. The model achieved an accuracy of 76% on the validation set. The model was trained on a set of approx 1700 images and around 300 images were taken for validation purpose. The images were grayscale and had shape of 96 X 96.

## Model Performance on real world data
Predictions for some real world color images.