Hello! I'm Isaac Corley

I’m a Machine Learning Research Scientist with a Ph.D. in Electrical Engineering from University of Texas at San Antonio (UTSA) advised by Paul Rad (Peyman Najafirad). I’m most passionate about Multimodal Learning and 3D Computer Vision particularly applied to Geospatial and Remote Sensing problems.

I’m excited about open-source software and currently maintain the TorchGeo and TorchSeg libraries.

I’ve worked as a full-time and contract ML Scientist/Engineer for several years at various companies where I’ve mainly applied computer vision methods to drone, aerial, and satellite imagery. In previous years I was heavily involved in machine learning applications of signal processing, cybersecurity, and biomedical sensors.


News

Publications

Depth Any Canopy: Leveraging Depth Foundation Models for Canopy Height Estimation

Depth Any Canopy: Leveraging Depth Foundation Models for Canopy Height Estimation

ECCV 2024 CV4E Workshop

We present a method for efficiently adapting monocular depth estimation models for tree canopy height estimation, providing superior or comparable performance to state-of-the-art methods with significantly less training data.

Estimating Earthquake Magnitude in Sentinel-1 Imagery via Ranking

Estimating Earthquake Magnitude in Sentinel-1 Imagery via Ranking

ECML-PKDD 2024 MACLEAN Workshop

We present a novel approach for estimating earthquake magnitude from Sentinel-1 imagery using a combination approach of regression and ranking based on earthquake magnitudes, significantly outperforming traditional regression only based methods.

Barely-Visible Surface Crack Detection for Wind Turbine Sustainability

Barely-Visible Surface Crack Detection for Wind Turbine Sustainability

IROS 2024

We present a novel dataset for evaluating the performance of surface crack detection models on wind turbine blades under challenging conditions with barely-visible cracks.

A Change Detection Reality Check

A Change Detection Reality Check

Isaac Corley, Caleb Robinson, Anthony Ortiz
ICLR 2024 ML4RS Workshop

We present an analysis of the current state-of-the-art in change detection literature. We find that a simple baseline of U-Net, an architecture from 2015, is still a top performer on several benchmarks and consistently outperforms many recently proposed methods.

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

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

IGARSS 2024

We introduce a novel remote sensing dataset for evaluating a model's ability to learn long-range spatial dependencies in aerial imagery by performing road extraction while containing large gaps occluded by tree canopy.

ZRG: A Dataset for Multimodal 3D Residential Rooftop Understanding

ZRG: A Dataset for Multimodal 3D Residential Rooftop Understanding

WACV 2024

We present a novel dataset for 3D understanding of roofs containing imagery, digital surface models, and 3D roof geometries captured from over 20k residential inspections.

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

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

CVPR 2024 PBVS Workshop

We perform a fair and large-scale evaluation of remote sensing foundation models and find chip/tile size and normalization preprocessing to be paramount for achieving peak performance for each individual pretrained model.

SSL4EO-L: Datasets and Foundation Models for Landsat Imagery

SSL4EO-L: Datasets and Foundation Models for Landsat Imagery

NeurIPS 2023

We introduce SSL4EO-L, the first ever dataset designed for self-supervised learning for Earth Observation for the Landsat family of satellites.

Solar Panel Mapping via Oriented Object Detection

Solar Panel Mapping via Oriented Object Detection

ICLR 2023 Tackling Climate Change with Machine Learning Workshop

We detail Zeitview's deployed end-to-end deep learning framework for detecting solar panels using rotated object detection architectures in large scale solar farms.

TorchGeo: Deep Learning with Geospatial Data

TorchGeo: Deep Learning with Geospatial Data

ACM SIGSPATIAL 2023

We introduce TorchGeo, a Python library for integrating geospatial data into the PyTorch deep learning ecosystem.

Supervising Remote Sensing Change Detection Models with 3D Surface Semantics

Supervising Remote Sensing Change Detection Models with 3D Surface Semantics

Isaac Corley, Peyman Najafirad
ICIP 2022

We propose Contrastive Surface-Image Pretraining (CSIP) for joint learning a latent space which extracts surface level features from optical RGB imagery which we show improves the performance in surface relevant tasks e.g. building segmentation and change detection.

Self-Supervised Representation Learning Enhances Broad Area Search in Multi-Temporal Satellite Imagery

Self-Supervised Representation Learning Enhances Broad Area Search in Multi-Temporal Satellite Imagery

IGARSS 2022

We propose pretraining on satellite image time-series (SITS) using the Contrastive Predictive Coding (CPC) self-supervised learning method which dramatically improves the performance of Broad Area Search (BAS) tasks in limited labeled data settings.

Destruction of Image Steganography using Generative Adversarial Networks

Destruction of Image Steganography using Generative Adversarial Networks

ArXiv 2019

We propose a Generative Adversarial Network (GAN) based method we coin Deep Digital Steganography Purifier (DDSP), which is optimized to remove steganographic content embedded in image pixels without compromising the perceptual quality of the original image.

DomainGAN: Generating Adversarial Examples to Attack Domain Generation Algorithm Classifiers

DomainGAN: Generating Adversarial Examples to Attack Domain Generation Algorithm Classifiers

ArXiv 2019

We propose DomainGAN which generates adversarial domains with near identical characteristics to benign domains and which greatly evades SOTA DGA classifiers.

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

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

Isaac Corley, Yufei Huang
IEEE Biomedical & Health Informatics (BHI) 2019

We train a Generative Adversarial Network (GAN) to upsample the spatial resolution of EEG datasets by generating realistic signals, eliminating the need for expensive EEG hardware.