Home

Hi! I am a researcher at IGN, the French National Geographical Institute, in the computer science department MATIS. The focus of my research is to exploit the graph structure of spatial/temporal data to improve the precision and speed of learning methods.

During my PhD I studied learning and optimization structured by large graphs (total variation, graphical models, etc). I now work on remote sensing application such as  semantic semantization of 3D point clouds, satellite hyperspectral imagery, and surface reconstruction. I have also developed an interest for structured deep models, such as graph convolutions, recurrent neural network, and attention-based encoders.

See my academic_CV for more details (updated in Feb 2020).

UPCOMING PRESENTATIONS:
– June 2020: Satellite Image Time Series Classification with Pixel-Set Encoders and Temporal Self-Attention, CVPR 2020 (oral)
– June 2020: Tutorial on Deep Learning for Remote Sensing at ISPRS2020, Nice
– March 2020: Tutorial on Deep Learning for Remote Sensing at EduServ-EuroSDR, Champs-Sur-Marne
– March 2019: Talk at QuantCube Technology: Recent Advances in Large-Scale Learning for Remote Sensing Applications

RECENT TALKS: (contact me if you want the presentations)

– December 2019: Introduction to Machine Learning and Deep Learning at NIBIO, Norway.
– Decembre 2019: Keynote Speech on Deep Learning for 3D Point Cloud Segmentation at EuroSDR Workshop on Point Cloud Processing (JNRR), Stuttgart
– November 2019: Talk at Valeo Labs: Deep Learning for 3D Point Clouds
– October 2019: Keynote Speech on Deep Learning for 3D Point Cloud Segmentation at Journées Nationales de la Recherche en Robotique (JNRR), Vittel
– June 2019: CVPR, Point Cloud Oversegmentation with Graph-Structured Deep Metric Learning
– June 2019: ISPRS Geospatial week,  University of Twente, Netherlands, Tutorial on Deep Learning for Point Cloud Semantic Segmentation
– June 2019: EduServ remote course on Deep Learning for Remote Sensing
– May 2019: Facebook AI Research, Exploiting Graph Sparsity
– May 2019: Université de Montpelier, Cut Pursuit for Optimizing with Graph-Structured Regularization
– April 2019, Champs-sur-Marne, Co-organiser of the 28th Conference on IGN Research
– 2 Avril 2019, Univ. Paris-Est, Structured Deep Learning for 3D Point Cloud Semantic Segmentation
March 2019, Champs-sur-Marne, Deep Metric Learning on 3D Point Clouds
– March 2019, Barcelona, Spain, EduServ, Deep Learning for Remote Sensing
– February 2019: Kigali, Rwanda, Teaching assistant at African Master of Machine Intelligence, Graphical Models with Guillaume Obozinski
– December 2018: University of Erlangen, Germany, Deep Metric Learning on Point Clouds
–  August 2018: Optimization in Image Analysis Summer School, Copenhagen
– July 2018: ICML2018, Stockholm: Cut-pursuit algorithm for Regularizing Non smooth Functionals with Graph Total Variation
– June 2018 : RFIAP2018, ENSG: Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs
– June 2018: CVPR2018, Salt Lake City : Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs (Poster)
June 2018 : SIAM symposium in Bologne, Graph Techniques for Image Processing
– June 2018 : NoMADS, Politecnico di Milano, Nonlocal Methods for Imaging
May 2018 : FOSS-4G, ENSG: presentation of the SuperPointGraph repository

NEW: Our newest paper on Sattelite Time Series has been acccepted at CVPR 2020 for an oral! Check the paper and the code. Comes with a new benchmark too.

NEW: Cut Pursuit is now fully parallel! Presented at ICML2019 Graph Reasoning Workshop, and faster than ever thanks to Hugo Raguet’s amazing implementation.

NEW: Our supervized partition scheme improve SPG’s performance by 6 points, and has been accepted to CVPR2019. See code and paper.

NEW: the new version of cut pursuit can now handle nonsmooth terms, and is considerably optimized (speed x1000 compared to state-of-the art for EEG inverse problems). The preprint is ready, the code as well!

NEW: our new framework superpoint graph, with Martin Simonovsky, is first in the Semantic3D challenge by 11.8 and 8.8 mIoU points. We also lead the S3DIS challenge by 12.4 mIoU  points. Check out the code here.

STUDENTS:
I currently co-advise 4 PhD students:
Stéphane Guinard
Mohamed Boussaha
Vivien Sainte Fare Garnot
Raphael Sulzer

I teach machine learning at master DesiGéo and master PPMD at ENSG as a teaching assistant, and in the machine learning course et ENCP.

I review for ICML, NIPS, ICCV, Mathematical Reviews, IJDSA, ANR, ISPRS among many others.