We are looking for an experienced Geospatial AI and Computer Vision expert to review the requirements for a Python-based tool that can automatically extract individual building and tent footprints from very-high-resolution satellite imagery.
The project will involve satellite imagery from Pléiades Neo at 30 cm resolution, as well as SkySat and Pléiades imagery at approximately 50 cm resolution. Existing imagery and manually validated footprints may be available for training, subject to a data-quality assessment.
The expected output is individual object polygons and object counts that can be exported as GeoJSON or shapefiles and used directly with ArcGIS.
The intended deployment environment is an AWS g5.4xlarge instance with an NVIDIA A10G GPU. The current environment is Windows-based, although Linux can be considered if technically justified.
The target processing volume is approximately 100 km² within one day.
No specific machine learning model or architecture has been selected yet. We are looking for an independent technical assessment based on practical experience to determine the most suitable approach for the project.
Relevant experience includes building or shelter extraction from satellite or aerial imagery, training and fine-tuning segmentation models, Python-based deep learning, geospatial data processing, and deployment. Experience converting model predictions into accurate and usable GIS polygons is particularly valuable.
The initial engagement will focus on reviewing the requirements, assessing the available data, recommending an appropriate technical approach, and providing practical guidance for implementation.
Please include a brief description of relevant projects you have worked on, your specific contribution, your availability, and your hourly rate.
