Description as a Tweet:

We wanted to convert our phones into a pokedex for animals. Point and recognize the creature in front of you with its species.


Our childhood and of course Pokemon

What it does:

As of now, we were able to run an object detector which given an image recognizes the object present in it. It outputs a probability score from a list of classes that could be assigned to an image.

How we built it:

We used AWS Sagemaker to train an image classifier. The platform had a built in image classification algorithm which we used to train on the caltech 256 dataset. We came across the iWildcam 2019 dataset to train an animal classifier but couldn't get to training one on it in the timeframe. Our learning curve to use AWS Sagemaker was steep and we encountered numerous issues to get the training done.

Technologies we used:

  • Python
  • AI/Machine Learning
  • Misc

Challenges we ran into:

Lack of GPU instances to train faster. First exposure to AWS Sagemaker made it difficult for us to figure out the concepts of buckets, notebook instances and compute instances.

Accomplishments we're proud of:

Finally being able to train a model on the cpu that does reasonably well to identify objects

What we've learned:

Technical skills - AWS, Pytorch, Hyperparameter tuning
Life skills - Patience, How to handle Disappointment

What's next:

Train an animal classifier

Built with:

Python, Caltech dataset, AWS sagemaker resources

Prizes we're going for:

  • Best Documentation
  • Best STEM Hack
  • Best AWS Hack
  • Funniest Hack
  • Best AI/ML Hack

Team Members

Shanu Vashishtha
Mayank Jha

Table Number

Table 47