Alex Stoken

Retreival based approach to astronaut photography localization.

Localization of nighttime astronaut photos by pairwise comparison with harmonized streelight and satellite data.

Pixel-wise matching of Astronaut photographs by way of iterative homography refinement.

Learn pixel-wise terrain roughness directly from image data - no DEM needed!

Geolocate astronaut photography of Earth via pairwise matching with reference satellite imagery.

Search for bounder+shadow to identify small boulders in low solar incidence angle regions (like the lunar South Pole!)

Pretraining a CNN on low level, structural vision tasks (edge detection, depth estimation) provides benefitial initialization for task-based policy learning.

Finding optimal cuts to seperate signal from background in the 4top-quark search at the ATLAS experiment.

Detect and track objects in and around the ISS to enhance operations and mission analysis. Built with very little training data.

Detect spacesuit joints for downstream pose analysis in NBL and EVA situations. Trained on simulated data only.

See the world through the lens of an astronaut inside the International Space Station! Explore, view, and interact with georeferenced astronaut photgraphy.

Unified API for 30 (and growing!) pairwise image matching models.

Methods for detrending and profiling craters in the South Pole, plus an analysis of small scale (<40m) crater morphology.

Recalibrated accelerometers in rover navigation systems can measure slight changes in gravitational acceleration, which can be used to map lateral variations in subsurface density.