Instance segmentation models for identification of recyclables on conveyor belts
Instance segmentation models for identification of recyclables on conveyor
belts.
## Code Structure
Note that these are demo models, reach out to
This is an implementation of Mask RCNN based on Python 3 and Tensorflow 2.x. The
waste-innovation-external@google.com for updated versions of the models.
model generates bounding boxes and segmentation masks for each instance of an
object in the image. The repository includes :
* Source code for training a Mask RCNN model.
## Overview
* Inference code
* Pre-trained weights for inferencing
* Docker to deploy the model in any operating system and run.
* Jupyter notebook to visualize the detection pipeline at every step.
* Evaluation metric of the validation dataset.
* Example of training on your own custom dataset.
The code is designed in such a way so that it can be extended. If you use it in
Mask RCNN is a state-of-art deep learning model for instance image segmentation,
your research or industrial solutions, then please consider citing this
where the goal is to assign instance level labels ( e.g. person1, person2, cat)
repository.
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.
## Pre-requisites
## Model Categories
## Prepare dataset
- Material model - Detects the high level category (e.g. plastic, paper, etc)
of an object according to its material type.
- Material Form model - Detects the category of the of an object according to
its physical product formation (e.g. cup, plate, pen, etc).
- Plastic model - Detects the category of a object according to its plastic
types (e.g. HDPE, LDPE, etc)
## Setup virtual systems for training
> The goal to develop these models is to bring transparency & traceability in
the world of waste recycling.
### ***Start a TPU v3-32 instance***
## Model paths in GCP buckets
- [x] Set up a Google cloud account on GCP
| Model categories | Model backbone | Model type | GCP bucket path |
- [x] Go to the cloud console and create a new project.
| ------ | ------ | ----- | ------ |
- [x] While setting up your project, you will be asked to set up a billing
| Material Model | Resnet | saved model & TFLite | [click here](https://storage.googleapis.com/tf_model_garden/vision/waste_identification_ml/material_model.zip) |
account. You will only be charged after you start using it.
| Material Form model | Resnet | saved model & TFLite | [click here](https://storage.googleapis.com/tf_model_garden/vision/waste_identification_ml/material_form_model.zip) |
- [x] Create a cloud TPU project
|Plastic model | Resnet| saved model & TFLite | [click here](https://storage.googleapis.com/tf_model_garden/vision/waste_identification_ml/plastic_types_model.zip) |
Light weight deep learning models for instance image segmentation.
## Overview
Mask RCNN is a state-of-art 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 model - Detects the high level category (e.g. plastic, paper, etc) of an object according to its material type.
- Material Form model - Detects the category of the of an object according to its physical product formation (e.g. cup, plate, pen, etc).
- Plastic model - Detects the category of a object according to its plastic types (e.g. HDPE, LDPE, etc)
> The goal to develop these models is to bring transparency & traceability in the world of waste recycling.
## Model paths in GCP buckets
| Model categories | Model backbone | Model type | GCP bucket path |
| ------ | ------ | ----- | ------ |
| Material Model | Resnet | saved model & TFLite | [click here](https://storage.googleapis.com/tf_model_garden/vision/waste_identification_ml/material_model.zip) |
| Material Form model | Resnet | saved model & TFLite | [click here](https://storage.googleapis.com/tf_model_garden/vision/waste_identification_ml/material_form_model.zip) |
|Plastic model | Resnet| saved model & TFLite | [click here](https://storage.googleapis.com/tf_model_garden/vision/waste_identification_ml/plastic_types_model.zip) |
