Building a Dog-Breed Classifier with Transfer Learning (AWS x Udacity)
This project was part of the Future AWS AI Scientist program.
The goal was to classify dog breeds using a pre-trained convolutional neural network.
Approach
I started with a ResNet-based model and fine-tuned it:
- Pre-processed images
- Augmented the dataset
- Froze base layers
- Trained final layers
- Evaluated on a validation set
Results
- Strong accuracy
- Improved generalization with augmentation
- Clear confusion matrix patterns for similar breeds
What I Learned
This project strengthened my understanding of:
- Transfer learning
- Training strategies
- Debugging CV pipelines
It was a great introduction to production-level computer vision on AWS.