Skip to content

The app takes image input from the user and accurately classifies the landmark in the image. It gives the top 5 possible landmark names with the probability.

License

Notifications You must be signed in to change notification settings

GauravG-20/Landmark-Image-Recognition-App

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

33 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Landmark Image Recognition App

Welcome to the Landmark Image Recognition App project! In this project, I have developed an application that can predict the most likely locations where user-supplied images were taken. The app uses Convolutional Neural Networks (CNNs) to analyze and classify landmarks in images.

Project Overview

The goal of this project is to create an app that can identify landmarks in user-uploaded images. Landmark recognition can be valuable for photo-sharing and storage services, as it allows them to associate location data with images, improving the user experience. This is especially useful when images lack location metadata, which can happen if the camera does not have GPS or for privacy reasons.

The app takes any user-supplied image as input and suggests the top k most relevant landmarks from a set of 50 possible landmarks from around the world.

Features

⚡Multi Label Image Classification
⚡Cutsom CNN
⚡Transfer Learning CNN
⚡PyTorch

Project Instructions

Development Environment

I have worked on this project in the Udacity Project Workspace, utilizing the provided GPU to speed up computations. The environment was already set up, including the starter code, which allowed me to focus on building the project.

Project Development

I followed the project instructions step by step:

  1. Building CNN from Scratch: I started by working on the cnn_from_scratch.ipynb notebook. In this notebook, I built a CNN model from scratch, trained it on landmark images, and evaluated its performance.

  2. Transfer Learning: The next step was to use transfer learning to improve model performance. I worked on the transfer_learning.ipynb notebook, where I fine-tuned a pre-trained model (ResNet-50) for landmark recognition.

  3. App Development: Finally, I created the Landmark Recognition App in the app.ipynb notebook. This app accepts user-supplied images and predicts the most likely landmarks based on the trained model.

Evaluation

To ensure the project met the requirements, I reviewed the CNN project rubric provided by Udacity and self-evaluated my project against it. I made sure that all criteria in the rubric were met to the best of my abilities.

Dataset Information

The landmark images used in this project are a subset of the Google Landmarks Dataset v2.

Project Results

Training Graphs

Here are some training graphs showing the model's performance during training.

CNN from Scratch Training Loss and Accuracy

CNN Training Loss and Accuracy

Transfer Learning Training Loss and Accuracy

Transfer Learning Training Loss and Accuracy

Deployed App

I have deployed the Landmark Recognition App, and it is accessible via the following link: Landmark Recognition App.

Conclusion

This project allowed me to gain valuable experience in building a real-world image recognition application using deep learning techniques. I learned how to develop CNN models from scratch, apply transfer learning to improve performance, and create a user-friendly app for landmark recognition. I also improved my skills in data preprocessing, model training, and evaluation.

If you have any questions or would like to explore the project in more detail, please feel free to reach out. Thank you for reviewing my Landmark Image Recognition App project!

License

MIT License

Get in touch

email linkedin

About

The app takes image input from the user and accurately classifies the landmark in the image. It gives the top 5 possible landmark names with the probability.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published