PolyFier

• 

  Sample photos loaded into cloud bucket.

• 

  Selecting labels for each sample photo.

• 

  Model begins training, estimated at two hours with 16 nodes.

• 

  Testing complete.

• 

  Logo-less aluminum can misses the mark.

• 

  White Stainless Steel bottle performs well.

• 

  Glass was not in the sample data, so it may be misinterpreted, however it is still accurate.

• 

  Branded soda can performs well.

• 

  Speaker shows some shortcomings of model, as the speaker falls below the standard for a mixed product.

• 

  Paper TO-GO box shows hole in logic for the model, where more sample data and "composting" parameter would improve real-world accuracy.

• 

  This too tested how geometry and texture affect output of model, and it seems to be accurate.

• 

  Similar result to white polystyrene TO-GO box, showing an possible area of improvement.

Sustainability is the future.

Usage:

Take a photo of a bottle and determine how much can be recycled.

How I built it:

Built a dataset to segregate mixed and fully recyclable bottles.

What did I do? Everything.

I have deprecated programming skills so this is a full on from-the-ground-up effort to learn Python and apply the Google Cloud Platform's AutoML API, make a dataset, test it.

Accomplishments:

Worked with Google Cloud services and can better apply features in future projects.

What I learned:TBA

What's next for PolyFier:

Integration into a mobile app for Android/iOS, add locale support to consider the ability of local facilities to recycle item in question.

Built With

• automl
• google-cloud