SBIR/STTR Award attributes
Deep learning-based computer vision models have been proven to produce highly accurate predictions for a variety of perception problems across numerous industries. This is largely due to the introduction of convolutional neural networks (CNNs), which significantly reduced the number of model parameters needed to process an image and explicitly utilize the spatial relationships between learned features. CNNs can be applied to numerous tasks such as image classification, scene segmentation, and object detection. While these networks are a powerful tool for enhancing the autonomy of a system, they typically require a well curated dataset of representative examples to learn from and expert knowledge to setup the training successfully. The proposed solution will address some of the challenges associated with using CNNs by making the process of training the model significantly easier through a user-friendly task-based interface, thereby reducing the need for an expert to complete the setup work.
