Integration of AWS and Roboflow Mask R-CNN Model for a Fully Cloud-Based Image Segmentation Platform

This study shows the integration of AWS to Roboflow through the Roboflow API. This will allow the utilization of the machine learning models within the Roboflow landscape thereby removing the need to reside the model within the AWS cloud. The developed Mask R-CNN model was able to achieve an overall recall of 81%, an average precision of 82%, an accuracy of 85.86%, and a mAP of 80.1%, with an average processing time of 1.5 minutes per image. 6,955 images were obtained from various sources such as manual on-site capture and online sources that were subjected to augmentation techniques resulting in a final dataset of 15,039 images. This was split into 80-10-10 train, test, and validation set for the model. This output aims to provide a framework that can help future researchers automate a cloud-based image segmentation model by integrating AWS and Roboflow.