Sebastian Gerard

I'm a fourth year PhD student, working with Josephine Sullivan. I am working at the intersection of computer vision and remote sensing.

At NeurIPS 2023, we recently published a remote sensing dataset for predicting wildfire spread from satellite images. My original goal with starting this PhD was to contribute to the global fight against climate change, so I am quite happy to be actively working towards this goal now.

• NeurIPS page, including paper
• Dataset download
• Code for baselines
• Code to create the dataset

Previously, I investigated a problem that could occur in certain domain-specific methods of self-supervised pre-training for remote sensing, that had been ignored in the existing literature. I wanted to quantify how big of a problem it could become, to determine whether practitioners need to take counter-measures. They do not.

• ArXiv link
• Twitter thread

At the beginning of my PhD work, we wanted to work with the xView2 / xBD dataset. After some failed attempts by others, I managed to simplify the highly engineered winning solution of the xView2 competition, while retaining most of the performance. I presented the results at the Living Planet Symposium 2022 in Bonn. We continued to work on this topic with Paul Borne-Pons during his research internship at KTH, during which he found some interesting details about the dataset composition and how this influences generalization. The corresponding preprint will be submitted to ArXiv.

• Twitter thread with LPS22 poster

Social links:

• Twitter
• LinkedIn