Anthony Fuller, “Why you shouldn’t trust results tables in remote-sensing-foundation-model (RSFM) papers”

Abstract: RSFMs are useful for all sorts of geospatial tasks. There are several dozen papers introducing new RSFMs that are trained with new self-supervised learning (SSL) algorithms. Yet we know very little about SSL for RS because the different models vary in their training data, training schedules, input and output modalities, and model architectures (despite being "ViT-B"). So we should standardize and start learning!

Bio: Anthony is a 3rd / final-year Ph.D. student at Carleton University / Vector Institute supervised by Jim Green (from Carleton) and Evan Shelhamer (from UBC / Vector). He is interested in deep learning, especially transformers and self-supervised learning. He has developed remote-sensing foundation models, e.g. CROMA and Galileo during his master's and Ph.D.