A small project made during the 2022 Hack112 Hackathon that uses a PyTorch machine learning model to categorize a given piece of trash captured by the webcam into either compost, recycling, or landfill.
With the advancements of technology and computer science, algorithms around automation have become more and more advanced. By streamlining the process of categorizing waste in a way that makes our everyday environments more sustainable, we effectively strive to create technology for social and environmental good.
- Orelia Pi (CS)
- Anna Shi (IS)
- Tim Wang (IS)
- Suanna Zhong (BXA)
All these libraries and modules can be downloaded using pip:
- numpy==1.23.4
- opencv_python==4.6.0.66
- Pillow==9.3.0
- pyscreenshot==3.0
- requests==2.25.1
- torch==1.13.0
- torchvision==0.14.0
There is no additional installments needed to run this application beyond downloading the necessary packages and libraries. To run the application in its full form, run main.py.
https://www.youtube.com/watch?v=jsXSy6KTIBw&ab_channel=TimothyWang
Given the dataset provided by Stanford containing 2527 images of assorted waste, categorized into glass, paper, cardboard, trash, metal, and plastic, we used a pretrained model from https://github.com/13w13/AI-Waste-Sorting-Web-App-Pytorch to output a predicted type of waste. After sorting through many different models, we found that this one was the most streamlined and applicable.
Given the short amount of time and limited previous experience regarding these libraries, we were limited to using a lot of borrowed concepts. If expansions were to be put into place, we would implement our own model and sample data in order to categorize not only the given six categories, but also food waste.