Visual Search – Landmark Identification

This project implements a visual search system for identifying landmarks from images. It was developed as part of a computer vision assignment using a dataset adapted from the Google Landmarks Dataset.

---

📊 Project Description

The goal is to determine whether two images belong to the same landmark or, given a query image, retrieve the k most similar images from a reference set.

The dataset contains 100 images across 20 landmarks (e.g., Sydney Opera House, Golden Gate Bridge) with variations in scale, pose, and lighting.

Two main tasks were completed:

Task 1 – Visual Search with Pretrained CNN

• Compute embeddings for query and reference images using a pretrained CNN.
• Measure similarity with a distance metric.
• Rank the k most similar images.
• Compute retrieval performance metrics.

Task 2 – Learning an Embedding

• Choose a CNN architecture (same as Task 1).
• Train with contrastive loss or triplet loss using the PyTorch Metric Learning library.
• Evaluate retrieval results and compare against Task 1.

---

🛠️ Requirements

Install dependencies with:

pip install -r requirements.txt

📈 Results

• Retrieval performance with pretrained CNN embeddings
• Improved results after training custom embeddings with contrastive/triplet loss