From Munch to Hunch: AI Classifies Your Waste at Lunch

By Karly Hou, Shruthi Jaganathan, Isaac Wilcove
Compost, recycle, or non-compost? The debate rages in your mind as you pace back and forth with your coffee cup in hand. With the help of the Green Machine, developed by three high school interns at NVIDIA this summer, classifying your waste has never been easier.

The three students, Karly Hou from Gunn High School, Shruthi Jaganathan from Cupertino High School, and Isaac Wilcove from Mountain View High School, hope their product can significantly help NVIDIA with its “go green” initiative.
Inside NVIDIA’s cafeteria, employees can place a tray of items on the cart and the Green Machine will use deep learning and a Jetson TX2-powered camera to identify each item and project color-coded bounding boxes on them. This project allows NVIDIANs to easily visualize how their waste should be sorted — compost, recycle, non-compost, and reusable.

The multi-class network, EnviroNet, was trained from SSD MobileNet V1 on an NVIDIA Tesla V100 GPU using the cuDNN-accelerated TensorFlow deep learning framework. The network is deployed on the Jetson TX2 using TensorRT for increased optimization. A projector is used to project the categorizations of the objects onto the tray.

The students chose to use a CSI-connected camera, taking advantage of the Jetson TX2’s powerful GPU for AI inference in order to cut latency and create near real-time AR-like interactivity.
The team hopes that the Green Machine’s ease of use will both educate and encourage people to sort their items correctly.
“Many people were surprised that the plastic cups and utensils should be composted,” the students said. “Every day we see how much waste is thrown away incorrectly, and we hope this device will make it easier and faster for people to sort correctly, helping NVIDIA do its part in reducing landfill buildup.”
Another key feature of the Green Machine is its transferability. Since many of the objects learned by the neural network are not specific to just NVIDIA, the modular hardware setup can easily be recreated and implemented in homes, offices, schools, stores, and more, with a few tweaks for any location-specific objects that need to be added to the network.