Home/Projects/dirt-road-segmentation

Dirt Road Segmentation Using Weakly Supervised Models

SHIPPED

Updated: Thu Aug 01 2024 00:00:00 GMT+0000 (Coordinated Universal Time) | Started: Fri Mar 01 2024 00:00:00 GMT+0000 (Coordinated Universal Time)

Computer vision project exploring dirt road segmentation without human annotation using SAM based area selection.

Notes

Overview

This project investigated automated segmentation of dirt road surfaces without manual labelling.

Description

Using a SAM based area selection workflow, reasonable segmentation performance was achieved without any human annotation. While not sufficient as a standalone solution, the approach proved useful at the time for rapid surface detection tasks.

Completed in mid 2024, the project could now be revisited using newer segmentation models to improve accuracy and enable real time operation.

Web App for Whole Body MRI Patient Onboarding and AI Reporting

Ongoing development of a secure web application for patient onboarding, internal workflows, and AI assisted reporting support for Whole Body MRI Australia.

Fake or Faker

Web based game challenging users to identify Faker among lookalikes and AI generated faces.