Skip to content

Research

Research and
Open Source.

Active research in foundation models, vision-language systems, and geospatial intelligence informs Bexar’s technical advisory.

Publications

Advisory informed by experience publishing peer-reviewed research at leading conferences, journals, and workshops—including NeurIPS, CVPR, ICCV, ICLR, WACV, and more. See the full list here.

First page of MIND the Gap: A Geographic Implicit Neural Representation with Adjustable Spatial Scale

First page of Fields of the Planet: Field Boundary Mapping Beyond 10m

First page of No One Knows the State of the Art in Geospatial Foundation Models

First page of Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?

First page of From Pixels to Patches: Pooling Strategies for Earth Embeddings

First page of Fields of The World: A Field Guide for Extracting Agricultural Field Boundaries

First page of Landsat-Bench: Datasets and Benchmarks for Landsat Foundation Models

First page of Aerial Infrared Health Monitoring of Solar Photovoltaic Farms at Scale

First page of FLAVARS: A Multimodal Foundational Language and Vision Alignment Model for Remote Sensing

First page of A Change Detection Reality Check

First page of Revisiting pre-trained remote sensing model benchmarks: resizing and normalization matters

First page of ZRG: A Dataset for Multimodal 3D Residential Rooftop Understanding

First page of Single-View Height Estimation with Conditional Diffusion Probabilistic Models

First page of Supervising Remote Sensing Change Detection Models with 3D Surface Semantics

First page of Destruction of Image Steganography using Generative Adversarial Networks

First page of DomainGAN: Generating Adversarial Examples to Attack Domain Generation Algorithm Classifiers

First page of Deep EEG Super-Resolution: Upsampling EEG Spatial Resolution with Generative Adversarial Networks

First page of Planetary Feature Fields are Scalable Earth Representations

First page of Chronosphere: Space-Time Tessellation of Local Climate Experts

First page of Infra-Bench CLS: A Global, Open-Source Benchmark for Critical Infrastructure Classification with Earth Observation Foundation Models

First page of Beyond Accuracy: Assessing Calibration of Geospatial Foundation Models and Their Sensitivity to Distribution Shifts

First page of HeatCast: A Benchmark for Neighborhood-Scale LST Forecasting across 124 U.S. Cities

First page of Earth Embeddings

First page of The first global agricultural field boundary map at 10m resolution

First page of PRUE: A Practical Recipe for Field Boundary Segmentation at Scale

First page of Advancing Earth Observation Through Machine Learning: A TorchGeo Tutorial

First page of Earth Embeddings as Products: Taxonomy, Ecosystem, and Standardized Access

First page of InspectVLM: Unified in Theory, Unreliable in Practice

First page of HydroChronos: Forecasting Decades of Surface Water Change

First page of TorchGeo: Deep Learning With Geospatial Data

First page of Depth Any Canopy: Leveraging Depth Foundation Models for Canopy Height Estimation

First page of Estimating Earthquake Magnitude in Sentinel-1 Imagery via Ranking

First page of Barely-Visible Surface Crack Detection for Wind Turbine Sustainability

First page of Seeing the roads through the trees: A benchmark for modeling spatial dependencies with aerial imagery

First page of Solar Panel Mapping via Oriented Object Detection

First page of SSL4EO-L: Datasets and Foundation Models for Landsat Imagery

Open-source leadership

Bexar’s founder maintains TorchGeo and Segmentation Models PyTorch, libraries for building and evaluating AI and Computer Vision systems. He also develops advanced production tooling for data pipelines and model deployment on high-performance compute clusters.

GitHub ↗ Research Blog ↗

Geospatial intelligence

Earth retrieval

Terrabit uses compact binary earth embeddings to retrieve similar satellite imagery in the browser.

Explore the proof of concept ↗

Coral atolls

Irrigated agriculture

Mining landscapes