Thomas Di Martino

PhD Student in AI & Remote Sensing

SONDRA @ CentraleSupélec/ONERA


As a French PhD student, I am passionate to whatever comes close to Artificial Intelligence & Earth Observation. Whether it is theoretical content with exploring state-of-the-art models or more concrete applicative programming with Jupyter Notebooks, I always find myself curious about what the world is up to !

Additionally, I am currently exploring the depth of SAR imagery, understanding its underlying concepts such as Polarimetric SAR or Interfometry.

You may also explore the personal projects I find myself doing on my free time hosted on this website. I have focused on exploring different facets of Deep Learning with applications in Multimodal Learning, Computer Vision or Automatic Feature extraction for Remote Sensing applications. All these projects’ repositories are hosted on Github.


  • Artificial Intelligence
  • Remote Sensing
  • Computer Vision
  • Earth Observation


  • PhD in Applied Physics, 2020-2023

    SONDRA Laboratory, CentraleSupélec, Gif-sur-Yvette, France

  • MSc in Artificial Intelligence & Multimodal Interaction, with Distinction, 2019-2020

    Heriot-Watt University, Edinburgh, Scotland

  • Engineering degree in Computer Science, 2017-2020

    EISTI, Cergy, France

  • BSc degree in Computer Science, 2015-2018

    Cergy-Pontoise University, Cergy, France



PhD Student in Applied Physics (application of Deep Learning to SAR Time Series)

SONDRA Laboratory | CentraleSupélec/ONERA

Sep 2020 – Present Paris, France

Working on problematics of change detection in SAR Time Series of forests with the help of Deep Learning methods.

SONDRA Laboratory is a laboratory mixing 4 entities: French ONERA, Supélec (known today as CentraleSupélec), the National University of Singapore and the DSO of Singapore.


Machine Learning and Computer Vision Intern


Apr 2020 – Aug 2020 Edinburgh, Scotland
Industrial Placement as part of the DataLab MSc program that led to the redaction of my Master Thesis.

  • Use of multimodal deep learning networks for duplicate product identification in a multi-retailer database.
  • Development of multimodal siamese networks with different losses evaluated and benchmarked (contrastive loss, triplet loss).
  • Analysis of sampling strategy: mini-batch hard sampling, semi-hard sampling, uniform-random sampling, mini-batch distance weighted negative sampling
  • Comparison with more traditional ML approach (PCA + Decision Tree + Handcoded textual features)

Deep Learning Intern


Apr 2019 – Sep 2019 Osny, France

As a deep learning Intern, I have trained, tuned and tested a model capable of doing building segmentation using satellite imagery. In this internship, I have tried multiple models (Mask RCNN, UNet, Deep UNet) and tried to take the best out of them all. The code was written using Keras with Tensorflow Back-End and was manipulated using a web-based RESTFul GUI with Flask and HTML5 technologies.

Also, multiple postprocessing technologies were considered and tried such as Logistic Regression (using scikit-learn) or Conditional Random Field (using pycrf).


Software Engineering Intern

Worldline ATOS

Jun 2018 – Sep 2018 Bezons, France
I have actively participated in the elaboration of a four tiers architecture implementing an Angular 6 Front-End, a Spring Boot middleware and J2EE back-office connected to COBOL programs using IBM JZOS technology.



Deep Similarity Learning & Siamese Networks

In this project, I explored deep similarity learning algorithms and their behaviour with different type of data (sequential data, spatial data, multimodal data). For each of these different modalities, I wrote 2 Medium articles detailing the retained method and providing my implementation.

Deep Reinforcement Learning projects

These 3 projects are implementations made for the udacity’s nanodegree program, all passed through a reviewer. They contain a small report, gathering my comprehension fo the algorithm as well as details on my implementation and my parameters.

Time Series Land Cover Challenge: a Deep Learning Perspective

In this project, I explored a Time Series of satellite images dataset by building different deep learning classifiers, finding inspiration in paper research in the field of Time Series classification.

Segmentation Models on artificial moon imagery

In this project, I trained 4 DL segmentation models on an artificial Lunar Dataset to see how they will perform on real moon images from Nasa.

Academic Publication

MSc Thesis: Multimodal Similarity Learning for Duplicate Product Identification

State of the art models for Similarity Learning are all based on Deep Learning architecture using Siamese Network [Gregory et al., 2015]. They define a feature extraction pipeline that creates a latent representation of input data. This embedding vector is semantically highly descriptive and can be used for the computation of distances between data records to measure similarity. While similarity learning is a popular topic, the combination of multiple modalities has not yet been attracted most of the attention of the field. In the context of Duplicate Product Identification, both the textual description of the product and their pictures can be used to make the similarity decision. This context of using data descriptors, e.g images and text, of different modalities require to rethink the concept of Siamese Network to perform multimodal similarity learning. In this work, multiple approaches have been explored: unimodal & multimodal Similarity Learning algorithms. The latter, combining embeddings across multiple modalities through gradient sharing method was proven to outperform any other combination of unimodal approaches through the use of N-Way & F-Beta Scoring. Furthermore, our analysis of the impact on the learnt features of the combination of multiple modalities has given insights on how they can collaborate to optimize the training function by detecting the most informative features for each modality. Comparing the weights of a multimodal siamese network to unimodal network helped to better evaluate cross-modality data profiles captured within the embeddings.

Blog Posts

Copernicus 2: The future of the Copernicus programme

The iconic European Programme that provided scientists with the Sentinel missions has something else up its sleeve and this for the …

REACTIV — Implementation for Sentinel Hub Custom Scripts Platform

REACTIV — Implementation for Sentinel Hub Custom Scripts Platform: Rapid and EAsy Change detection in radar TIme-series by Variation …

How to choose your loss when designing a Siamese Neural Network ? Contrastive, Triplet or Quadruplet ?

Deep Similarity Learning is the training of a deep learning architecture to learn to detect similarity and disimilarity between two …

Introduction to Deep Similarity Learning for sequences

Deep Similarity Learning is the training of a deep learning architecture to learn to detect similarity and disimilarity between two …

Time Series Land Cover Challenge: a Deep Learning Perspective

Time Series Satellite Imagery is the addition of a temporal dimension to Satellite Imagery. We see in this post how a 10-bands …


Winner of the Data Fusion Contest 2021 - Detection of Settlements without Electricity

My team (Maxime Lenormand, Elise COLIN KOENIGUER) tied for 3rd place at the Data Fusion Contest 2021.The challenge in question, involving the detection of settlements without electricity, aims to leverage multimodal and multitemporal remote sensing data, combining SAR & Optical data, for the greater good.For that task, we developed a custom Multi-Channel Deep Learning architecture that we will present during an invited session at IGARSS 2021, in Belgium.

Winner of 2 categories of the Sentinel Hub custom script competition 2020

Collaborative work realized by me, Elise COLIN KOENIGUER, Regis Guinvarc’h, and Laetitia Thirion-Lefevre with the implementation of REACTIV, a multi-temporal method for change visualization in SAR Time Series, has been announced as the winning submission of the Early Bird section of the Sentinel Hub custom scripts competition.

The Data Lab MSc Scholarship

I was offered a scholarship to pay for my tuition fees, as well as invitations to all DataLab events for my 2019-2020 year of study at Heriot-Watt University. Award winning candidates were chosen based on their resume as well as a statement of purpose.
The Data Lab organisation puts emphasis on making connections between Scotland’s finnest Data Scientist and data students. These events are opportunities to chat with professionals and exchange knowledge with other students.