This project aims to explore the effectiveness of a Siamese Neural Network (SNN) in developing a robust image retrieval model by leveraging image key points. Through rigorous experimentation and analysis, the project investigates the SNN's capabilities in capturing intricate image similarities and addresses the challenges associated with image retrieval tasks. The achieved performance demonstrates the SNN's potential in enhancing image retrieval accuracy and robustness.
The comprehensive research paper detailing the methodology, experiments, and findings of this project is included in the repository. Please refer to the Research Paper for an in-depth analysis of the project, including detailed descriptions of the methodologies employed, experimental setups, results, and discussions.
The project employs a Siamese Neural Network architecture, incorporating image key points for feature extraction and similarity measurement. The network undergoes comprehensive training using a dataset containing image pairs and corresponding relevance scores. Evaluation metrics such as precision at k and accuracy within a relevance score range are utilized to assess the model's performance.
Image Key-point Extraction:
Our Siamese Neural Network Model:
- The Siamese Neural Network (SNN) achieved a 10-precision of 2%.
- The SNN demonstrated an accuracy of 3% in classifying relevant images within a relevance score range of 0.1.
Image Query:
Images Retrieved:
In conclusion, our investigation into employing Siamese Neural Networks (SNNs) alongside keypoint extraction for image retrieval has revealed inherent flaws and limitations within this approach. Although the SNN displayed marginal effectiveness in identifying relevant images, the overall performance, coupled with its complexity, signifies that this model is not a viable solution for the information retrieval task at hand.
Moving forward, a more streamlined approach involving a simple SNN with convolutional layers is anticipated to yield superior results. This alternative strategy aims to capitalize on the power of convolutional layers for feature extraction while minimizing unnecessary complexities.
model.ipynb: Python notebook script containing the implementation of the Siamese Neural Network and the image retrieval process.process_data.ipynb: Python notebook script for preprocessing the dataset and extracting image key points.
This project is licensed under the MIT License - see the LICENSE file for details.