This project is one of the challenges in the online courses of Microsoft Azure Data Scientist Associate on Coursera.
Goal:
To create a convolution neural network for classifying different animal images.
Dataset:
From Microsoft and it is a simple dataset with only four types of animals.
Process:
- Review the images.
- Transform the images to tensor and perform normalization.
- Split the data for training and testing and load them to DataLoader.
- Build three convolution layers to extract features, followed by pooling layer and ReLu activation function.
- Drop some features to avoid overfitting.
- Evaluate the model with confusion matrix.
- Use the model to predict unseen images in the testing folder.
- These are implemented in both PyTorch and Tensorflow.