Data Sheets


Issue link:

Contents of this Issue


Page 0 of 1

data sheet FEATURES The DeepCore Suite, developed by Maxar, is an end-to-end artificial intelligence and machine learning (AI/ML) pipeline for self-service computer vision (CV) analytics. It evolved from Maxar's direct experience as an industry leader developing automated computer vision techniques for commercial and national technical means imagery. The U.S. government has Small Business Innovative Research (SBIR) data rights for DeepCore Suite, allowing cost-free and scalable use. What can DeepCore offer you and your team? DeepCore Suite augments data scientists' and analysts' workflows by allowing them to train, deploy and run CV models at scale and at the edge to deliver meaningful insights against some of our nation's hardest challenges. What does it do? DEEPCORE SUITE AI accessibility for anyone. Maxar data science teams worked with government and industry to build easy-access, mission-sensitive operational models. Customers don't need extensive data science experience to create training data, deploy models and manage large-scale CV model runs. The DeepCore Suite includes components to assist in these tasks. Capabilities with compatibility. DeepCore has deployed more than 100 models to detect 130+ object types using multiple machine learning models, frameworks and networks against multiple satellite, airborne, drone and terrestrial sources. It has also been deployed in commercial and government clouds, as well as bare metal and hybrid environments scaling from a machine with a Nvidia GPU that supports CUDA to an infinitely scalable cluster. With DeepCore, users can leverage the services and experience of the Maxar team or mix and match training data, models and visualization capabilities within DeepCore or other repositories and tools. ■ Powerful inference engine that can be miniaturized and scaled to address speed, size and complex use cases □ Deployed on one GPU box at the edge (laptop, drone, Worldview Legion satellite) and scales infinitely with minimal performance degradation across cloud, hybrid and on-premise GPU clusters □ Inference for speed size or complexity, including sensor to shooter, satellite to RGT, thousands of daily 50+ GB imagery strips, different imagery repositories, and such algorithm types and analysis use cases as broad area search and rare object detection ■ 100+ CV and ML models deployed in DeepCore to date via □ Frameworks including Caffe, Caffe2 via an Open Neural Network (ONNX) packaging format, TensorFlow and PyTorch via Torch Script □ Network types like canonical classifiers (Resnet), segmenters (Mask-RCNN) and object detectors (RetinaNet, Faster-RCNN, Yolo) ■ Broad support and interoperability with other external frameworks □ Broad direct framework support □ Improvements to interpret additional model types □ Ability to upload and deploy third-party models as zip files or ONNX Imagery for training data Imagery for User 2 3 5 Vision Training data catalog Model catalog Collection manager Inference server 4 1 Tagging & validation Synthetic Model development and training Imagery Services Verifi ed are added to training data to generate improved machine learning model

Articles in this issue

Links on this page

view archives of Data Sheets - DeepCore