Skip to content

olyapka84/Landmark-Classification-CNN-Project-for-Udacity

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 

Repository files navigation

Landmark Classification — CNN Project

A deep learning project for classifying landmarks in images using Convolutional Neural Networks (CNNs).

🎯 Project goal

Build and compare CNN models to automatically recognize famous landmarks from photos.
This helps infer the photo’s location when GPS metadata is missing — a task useful for photo storage and tagging services.


🧩 Project structure

File / Folder Description
cnn_from_scratch.ipynb Build and train a CNN from scratch
transfer_learning.ipynb Apply transfer learning (ResNet, VGG, etc.)
app.ipynb Deploy and test the best model
src/data.py Data loading and preprocessing
src/model.py Model architecture definitions
src/train.py Training logic
src/transfer.py Transfer learning setup
src/optimization.py Hyperparameter tuning
src/predictor.py Inference utilities
src/create_submit_pkg.py Script to package submission
src/helpers.py Helper functions
cnn_from_scratch.html HTML export
transfer_learning.html HTML export
app.html HTML export
README.md Project overview

🧠 What’s inside

  • CNN model built from scratch with PyTorch
  • Transfer learning model fine-tuned on the same dataset
  • Model comparison and evaluation
  • Deployment of the best-performing model in a simple app interface

📊 Results

  • Baseline CNN reached good accuracy
  • Transfer learning improved classification performance significantly (>90%)
  • The final model was exported via Torch Script and integrated into an app

🧰 Tech stack

  • Python
  • PyTorch / torchvision
  • numpy / matplotlib / tqdm / PIL
  • Jupyter Notebook

📚 Created as part of a deep learning course to practice CNN design, transfer learning, and deployment.

About

A deep learning project for classifying landmarks in images using Convolutional Neural Networks (CNNs).

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published