Garbage Classification

Inspiration

We were inspired by simply looking at a nearby recycling bin and seeing all of the improperly recycled materials. This lead us to think that if there was some type of identification software then maybe in the future garbage sorting could become automated.

What it does

Upon properly loading all of the code cells in Google Colab, a Gradio GUI will appear near the bottom of the page and you can select an image to upload. After pressing submit and waiting a few seconds, near the upper right side of the GUI a guess of the item's material will be made.

How we built it

We found a trash classification dataset on Kaggle that had 2000+ images with labels. Additionally, we found several methods online to train models for this dataset, so we closely followed one of those methods. This method used the ResNet50 architecture, which is famous due to its success in the ImageNet competition, as well as PyTorch to train and feed data into the model. After successfully training the model to have an accuracy of over 90%, we used Gradio to setup a demo for this model. This allowed users to input an image and output a corresponding label to that image.

Challenges we ran into

The biggest challenges we ran into were getting Gradio to properly mesh with our model, as all the inputs/outputs from our model prediction had to align perfectly with the inputs/outputs feed to and from Gradio.

Accomplishments that we're proud of

After hours of hard work and hand cramps, we were ecstatic when the program finally compiled and made a correct prediction.

What we learned

We inevitably learnt a lot about machine learning and how surprisingly useful the Google Colab platform is for these types of projects. We also learned about using the PyTorch framework