Identifying Bikers Distracted by Cellphones

After doing Week 2, I decided to try and build a classifier to identify bike riders who pose a danger to themselves and to others on the road by speaking or texting on their phones while riding.

You can see the the Jupyter notebook here: https://github.com/vikram-s-narayan/ridingWithCellphones

The problem is that I’m not able to get more examples of distracted bike (both bicycle and motorbike) riders. My dataset consists of two classes of photos:

  1. “OK” riders (i.e. those who are riding without simultaneously using their phones)
  2. “Not OK” riders (those who are texting or speaking while riding)

Each of the classes has about 100+ photos.

My question is as under:

The accuracy is currently at 80% (please see results below). Is there any way to improve this without finding further examples of distracted riders?

You may add more data augmentation instead of using only the defaults.

1 Like

Thanks @ptrampert. I’ll take a look at data augmentation.

Try giving object detection a look too, might help with the accuracy a little but will make it slower for inference and training