FDL Europe 2024

FDL Europe 2024 - Earth Systems Lab - 3D clouds using multisensors

I took part in the FDL-Europe research sprint 2024 and once again loved my experience. On this occasion aswell as continuing with the team-oriented and multi-disciplinary aspects of the research sprint I was able to learn best practices in terms of collaborative coding and learn state of the art deep learning techniques.

My team’s challenge consisted in:

1. Develop a self-supervision model that allowed us to exploit the full geostationary archive of images (>10TB). This model finds general purpose embeddings for 2d multispectral images.
2. Build a fine-tuning model to translate the embeddings found in (1) into a 3D cube of cloud reflectances.
3. Build a fine tuing model to translate the embeddings found in (1) into a 3D cube segmentation of cloud types.

The team consisted of 4 wonderful researchers:

1. Lilli Freischem
2. Stella Girtsou
3. Joppe Massant
4. Myseif

Four excellent faculty members:

1. Anna Jungbluth
2. Kyriaki-Margarita Bintsi
3. William Jones
4. Guiseppe Castiglioni

and a great advisers in Michael Eisinger and Emmanule Johnsson. I would like to thank our lovely supporting team of Catherine Ballarin and Cormac Purcell.

Here is a link to the youtube showcase presentation.

Here is a link to the neurips paper “3D Cloud reconstruction through geospatially-aware Masked Autoencoders”.

Here is a link to the project repository.