Data wrangling: A brief analysis of training data.
Classification- Damaged vs. Undamaged: Here I have classified whether a car is damaged or not using learners built on fastai libraries.
Classification- Damage type: Here I have categorized the damage based on its location.
Prediction from video: In this notebook I have created frames from a input video, classify each frame using the pretrained classifier, annotate each frame acoordingly, and finally make a video from those annotated frames. This would be useful for damage assessment, where a user would feed a video of a damaged car as input, and the damage location/extent would be predicted instantly. This can be very beneficial for insurance companies as well as the car owners.
The video link for damage prediction