Used the Facial Keypoint dataset from Kaggle which is data on several images and corresponding data points on those images.
The variables in the dataset are the locations of each keypoint
I have performed the following tasks on this dataset:
Task 1:
- Data Analysis and Visualisation on the dataset
- Converting the images into dimensions readable by the model and showing the images
Task 2:
- Building a CNN using Keras
- This model takes the image in the test data as an input and predicts keypoints
- Task 2a
- Used transfer learning to replace the custom CNN with ResNet architecture.
Task 3:
- Deployed this model using Flask and built an interface that takes an image and outputs the predicted values overlayed on the same image
- Deployed the model on Heroku
Task 4:
- Using the model (with a new dataset) to predict facial expressions