Defense Innovation Unit Experimental (DIUx) and the National Geospatial-Intelligence Agency (NGA) Satellite Imagery Dataset

xView features a diverse collection of small, rare, fine grained, and multi-type objects with bounding box annotation. (source: DIU)

xView features a diverse collection of small, rare, fine grained, and multi-type objects with bounding box annotation. (source: DIU)

March 13, 2019 | Source: Defense Innovation Unit (DIU), https://www.diu.mil/, 2018

Defense Innovation Unit Experimental (DIUx) and the National Geospatial-Intelligence Agency (NGA) released a new satellite imagery dataset to advance key frontiers in computer vision and develop new solutions for national security and disaster response.


The release was in support of the DIUx xView 2018 Detection Challenge: OBJECTS IN CONTEXT IN OVERHEAD IMAGERY

WHAT IS xVIEW?

xView is one of the largest publicly available datasets of overhead imagery. It contains images from complex scenes around the world, annotated using bounding boxes. The DIUx xView 2018 Detection Challenge is focused on accelerating progress in four computer vision frontiers:

  1. Reduce minimum resolution for detection
  2. Improve learning efficiency
  3. Enable discovery of more object classes
  4. Improve detection of fine-grained classes

xView follows in the footsteps of challenges such as Common Objects in Context (COCO) and seeks to build off SpaceNet and Functional Map of the World (FMoW) to apply computer vision to the growing amount of available imagery from space so that we can understand the visual world in new ways and address a range of important applications.

xView comes with a pre-trained baseline model using the TensorFlow object detection API, as well as an example for PyTorch. More details about the dataset and initial experiments can be found in our NIPS poster presented at the Machine Learning for the Developing World workshop.


Related Materials:

The xView dataset is large--1 million object images with 60 classes and a 0.3 meter resolution--and labeled using crowd-sourced annotations for bounding boxes and classes. The paper that DIUx included to accompany the release of the dataset, xView: Objects in Context in Overhead Imagery, discusses that the collection techniques by nature accepted a level of noisiness in the data.

DIU Paper: xView: Objects in Context in Overhead Imagery

Carnegie Mellon University article discussing the challenge and dataset: Deep Learning and Satellite Imagery: DIUx Xview Challenge

GitHub for the DIUx xView Detection Challenge (provides links to Baseline [baseline models, scoring, and inference] and Data Utilities [utilities for processing the xView dataset])

Communities: