Facial keypoints detection

Using Convolutional Neural Network (CNN) based model to train and infer the facial keypoints of images. The images showed two distinct groups one with upto 8 facial keypoints and the other with more than 8, and upto 30 keypoints. Using similar but seperate models for training and predicting facial keypoints of the two groups of images, in the end, the Root Mean Square Error was 1.719, low enough for our team to be placed 4th on the Kaggle's public leaderboard (post competition).