Assignment

Our assignment was to develop a machine learning model that could accurately classify images. We chose to classify Formula 1 cars by their respective teams. This was a challenging task, as the cars often have very similar designs.

Approach

In this project, we were a total of three students. We collected a diverse set of images of Formula 1 cars from five different teams: Mercedes, McLaren, Red Bull, Ferrari, and Renault. These images were collected from flickr using a web scraper, ensuring a high-quality diverse representation of each team's vehicles. This method allowed for gathering a vast range of images varying in angles, race conditions, and years, providing a rich dataset for model training. Then, we did transfer learning with the FastAI library. We chose a pre-trained ResNet50 model as our base and fine-tuned it on our dataset of F1 car images. We experimented with different learning rates and training techniques to optimize the model's performance.

Result

In the end, we were able to develop a model that achieved a high level of accuracy in classifying F1 cars by their teams. The model was able to generalize well to new images, and it was robust to variations in lighting and camera angles. We evaluated the model using a variety of metrics, including a confusion matrix, ROC curve, and AUC. The model showed high accuracy across all classes, with perfect classification for some teams.

My Input

I actively collaborated with my fellow team members throughout the project. My contributions included cleaning and preprocessing the dataset. Additionally, I played a role in implementing the transfer learning approach using the FastAI library, fine-tuning the pre-trained ResNet50 model on our specific dataset of Formula 1 car images.

What I learned

Through this project, I gained a deeper understanding of transfer learning and how it can be used to solve real-world problems. I am proud of the results we achieved, and I believe that this project has prepared me well for future work in the field of computer vision.