Bean Counter

Inspiration

As research scientists in the area, we are motivated to make a difference on a serious challenge.

What it does

We can separate healthy grains from contamination based on machine perception.

How we built it

We used unsupervised learning to pre-process the data, following the premise that photos are cheap, labels are expensive. We extract information by auto-encoding, and the embedding helps us to leverage carefully labeled data.

Challenges we ran into

The key challenge is the missing data. In particular, in the limited amount of time, we have to work with minimal labelling.

Accomplishments that we're proud of

We leveraged recent advances in the scientific community on a real-world problem. We built a light-weight, easy-to-use platform for labeling.

What we learned

Unsupervised learning is the future.

What's next for Bean Counter

Consolidate the code base and apply the techniques to new challenges.

Built With

• machine-learning
• python
• tensorflow