This project was done as a part CS-574 (Computer Vision using Machine Learning) course at IIT Guwahati.

1. Numpy
2. Python 3.6
3. Keras(with Tensorflow Backend)
4. Pandas
5. Scikit-Learn

We have tried out 4 types of Architecture namely :

1. RandomForest
   • Data was split into 23% test data and 77% training data.
   • RMSE = 2.72725802
2. Feed-forward Neural Network (with single hidden layer of 100 neuron)
   • Data was split into 23% test data and 77% training data.
   • RMSE = 2.98529003976
   • Some result on test data:
   • 
3. Convolutional Neural Network
   • Data was split into 23% test data and 77% training data.
   • RMSE = 2.5521943944
   • Following CNN architecture was used for model-3 and model-4 :
   • 
4. Convolutional Neural Network (with dropout on every layer)
   • Data was split into 23% test data and 77% training data.
   • Dropout probablity of 0.5 was used in model 4.
   • RMSE = 2.32341972693
   • Following is the comparison of model-3 and model-4 :
   • 

Our result showed that CNN(with dropout) performed best on facial keypoint detection task followed by CNN(without dropout), RandomForest and Simple Feed Forward Network respectively.

1. Abhinav Anshuman
2. Aditya Gaurav
3. Chirag Sodani
4. Rahul Kumar