models/official/projects/waste_identification_ml

CircularNet

Instance segmentation models for identification of recyclables on conveyor belts.

We provide retraining and fine-tuning utilities, but if you're interested in partnering more closely with us reach out to waste-innovation-external@google.com

Overview

CircularNet is built using Mask RCNN, which is a deep learning model for instance image segmentation, where the goal is to assign instance level labels (e.g. person1, person2, cat) to every pixel in an input image.

Mask RCNN algorithm is available in the TensorFlow Model Garden, which is a repository with a number of different implementations of state-of-the-art models and modeling solutions for TensorFlow users.

Model Categories

• Material Type: Identifies the material type (metal, paper etc) of an object. For plastic, resin types are also identified (HDPE, PET, LDPE, etc).
• Material Form: Categorizes objects based on the form factor (cup, bottle, bag etc)
• Example inference label: Plastics-PET_Bottle

Latest model

Single unified model that performs material type and form detections

Model categories	Model backbone	Model type	GCP bucket path
Material Type & Form	Resnet	saved model	click here

Full Documentation

The full documentation, covering everything from how to choose and install a camera to how to prepare and make use of the model is here. Below, we also provide a quicker guide for running inference using a GCP VM, assuming you already have a working camera taking pictures.

End to End Cloud Deployment Guide

End to end deployment involves three key steps:

1. GCP GPU VM creation

2. Code configuration

3. Results analysis

We will go through each one of them in details below

[A] Prerequisite - Create VM instance:

Create a Google cloud account and a T4 GPU enabled VM:

• Create VM in GCP Cloud

[B] Code Setup - Clone and start the pipeline

Run the following commands mentioned in each step on the SSH-in-browser window of your VM instance in Google Cloud

Step 1:

• Clone the repository

Step 2:

• Start the server

Step 3:

• Run the prediction Pipeline

For more details: Click Here

[C] Setup Dashboard - Visualize results

For reporting purposes and to analyze image categories, we need to set up and connect looker dashboard with BigQuery table:

• Prepare and analyze images

Authors and Maintainers

Umair Sabir - Primary developer Sujit Sanjeev - Product Manager Ethan Steele - Collaborator Vinit Ganorkar - Collaborator