Udacity Introduction to Deep Learning Nanodegree (Core #4) — Course Project
Build a landmark classifier that automatically predicts the location of images based on depicted landmarks. This solves the problem of tagging photos that lack GPS metadata on photo-sharing platforms.
├── src/
│ ├── data.py # Data loading, transforms, visualization
│ ├── model.py # Custom 5-layer CNN architecture
│ ├── optimization.py # Loss function and optimizer
│ ├── train.py # Training loop, validation, testing
│ ├── transfer.py # ResNet50 transfer learning setup
│ ├── predictor.py # TorchScript export and inference
│ └── helpers.py # Utility functions
├── cnn_from_scratch.ipynb # Part 1: Train CNN from scratch (≥50%)
├── transfer_learning.ipynb # Part 2: Transfer learning with ResNet50 (≥60%)
├── app.ipynb # Part 3: Export and deploy model
├── requirements.txt
└── README.md
50 classes of world landmarks. Place the dataset in landmark_images/ with train/, valid/, test/ subdirectories.
pip install -r requirements.txt
# Run notebooks in order: Part 1 → Part 2 → Part 3