Overview / Usage

The project is dedicated to developing a cutting-edge deep-learning model for object detection in autonomous vehicles. The primary objective is to ensure seamless navigation and prompt processing. The project involves training three distinct model architectures: YOLOv7, YOLOv5 and Faster R-CNN. To enhance the model's training and inference speed, advanced tools such as the Intel Extension for PyTorch and Intel Neural Compressor are effectively utilized. These tools optimize performance by capitalizing on hardware capabilities and compression techniques. Additionally, a user-friendly web interface is crafted using Streamlit, allowing for easy implementation of all models.

This project effectively tackles the crucial challenge of accurate object detection in autonomous vehicles, enabling them to make informed decisions and navigate with utmost safety. By harnessing the potential of deep learning models and real-time processing, the system substantially bolsters situational awareness and decision-making capabilities. The Intel tools play a pivotal role in optimizing training and inference speed, while the web interface streamlines model evaluation and visualization. The comprehensive approach adopted in this project contributes significantly to developing reliable object detection systems for autonomous vehicles, thereby facilitating their secure deployment in real-world production environments.

Methodology / Approach

Our methodology for developing the object detection deep learning model for autonomous vehicles follows a systematic approach, utilizing advanced technologies to effectively solve the problem. We incorporate multiple model architectures including YOLOv7, YOLOv5 and Faster R-CNN to explore their strengths and weaknesses. By leveraging the Intel Extension for PyTorch and Intel Neural Compressor we optimize training and inference speed through hardware capabilities and compression techniques. Adhering to industry standards and employing rigorous data preprocessing techniques, we ensure high-quality training data for accurate object detection. Our methodology emphasizes comprehensive testing and validation, evaluating model performance using standard metrics such as precision, recall, and mean average precision.

Technologies Used

I used Intel dev-cloud to train and test my model. Also, I implement Intel Extension for Pytorch and Intel Neural compressor to optimize my model performance on both the training and inference stages. I chose 3 model architectures such as YOLOv7, YOLOv5 and Faster-R-CNN and made a comparison between those models. After I trained my model Streamlit is used to Deploy my model as a Web Application.

Repository

https://github.com/rppadmakumar3/ObjectDetectionOneAPI