Clarifai Response to Topic Number: NGA 201-001 Phase II Rare Objects

Computer Vision AI models are trained on large datasets to understand imagery and assist with human analysis in thousands of use cases. This proposal is intended to enable the use of machine-learned computer vision to assist analysts in the location of rare objects even when there is insufficient real world labeled data for deep training. The generic problem of automatic recognition of objects from sensor data is already a multi-billion dollar issue in both the government and the private sector. While this project focuses on use of synthetic data of aerial imagery to generate AI models to find objects where there may not be sufficient training data, the techniques are useful in a broad range of use cases.