TorchGeo: Deep Learning with Geospatial Data

Event Type

Seminar/Symposium

Sponsor

New Frontiers Initiative

Date

Mar 15, 2022   10:00 am  

Speaker

Adam Stewart, NFI Fellow, Illinois Computer Science PhD Student

Registration

Free to attend – register here.

Views

118

Originating Calendar

NCSA-related events

New Frontiers Fellow and Illinois Computer Science PhD student Adam Stewart lead an exciting webinar on March 15 @ 10 a.m. to introduce folks to TorchGeo, an open-source Python library for integrating geospatial data into the PyTorch ecosystem. TorchGeo helps realize the potential of deep learning for remote-sensing applications by providing pre-trained models for multispectral satellite imagery, benchmarking and more.

Abstract Summary
Remotely sensed geospatial data are critical for applications including precision agriculture, urban planning, disaster monitoring and response and climate change research, among others. Deep learning methods are promising for modeling remote-sensing tasks, given the success of deep neural networks. However, the variance in data collection methods and geospatial-metadata handling makes applying deep learning methodology to remotely sensed data nontrivial.

TorchGeo is the first library to offer pre-trained models for multispectral-satellite imagery, allowing for advances in transfer learning on downstream remote-sensing tasks, with limited labeled data. TorchGeo creates reproducible benchmark results on existing datasets and benchmarks a proposed method for preprocessing geospatial imagery on the fly. Read the full abstract here.