The theme of Hack-ccelerate helped us discover this exciting project. We wished to explore Computer vision based projects, and hence we chose to create an automobile parts classifier using CNN.

Our deep learning model classifies 14 different types of automobile parts.

The model was built mainly using the tensorflow keras API.

The main challenge we faced was handling the data in such a way as to prevent overfitting. This was overcome using data augmentation and a proper CNN architecture, that was created with thorough experimentation.

We were able to achieve a good accuracy, without using any kinda of transfer learning. Our custom model was able to compete with already established tensorflow architectures.

Working on creating a custom CNN architecture helped us in creating a strong intuition about how different architectures perform, how different layers affect training, etc.

The next goal is to improve the generalisation of the model.

• Amruthaa S
• Grace Hephzibah M