I am an aspiring Computer Vision researcher, and I am extremely lucky to be working with Prof. Vincent Sitzmann as a visiting student in the Scene Representation Group at MIT CSAIL. In a few words, the vision that I pursue through my research is the development of Perception systems that can learn from minimal supervision by drawing inspiration from first principles.

In the past, I have been lucky to collaborate with and learn from fantastic researchers to start advancing towards this goal. I have spent a summer collaborating with Prof. Jia Deng in the Princeton Vision & Learning Lab, where I investigated 4D reconstruction from videos of highly-dynamic scenes. I have also spent a wonderful summer working with Prof. Mathieu Aubry in the Imagine team at ENPC, working on weakly-supervised learning of visual prototypes with applications to document analysis.


Mar. 2024 - The Learnable Typewriter - our paper on document analysis using weak or no supervision - has been accepted to ICDAR 2024!

Feb. 2024 - Excited to start my visiting stay with Prof. Vincent Sitzmann in the Scene Representation Group!

Sep. 2023 - I obtained the Master MVA from École Normale Supérieure Paris-Saclay!

Research experiences

Feb. 2024 -
Visiting student working with Prof. Vincent Sitzmann in the Scene Representation Group at MIT CSAIL.

May - Sep. 2023
Master's Thesis advised by Prof. Jia Deng in the Princeton Vision & Learning Lab.
Subject: 3D Vision, Scene Representation.

Apr. - Sep. 2022
Research internship advised by Prof. Mathieu Aubry in the Imagine team at ENPC.
Subject: Weakly-supervised methods for analysis of Text-line images.

May. - Aug. 2021
Research internship advised by Prof. Emmanuel Dellandrea at CNRS LIRIS.
Subject: Real-time Visual Scene Understanding for Robotics.


2022 - 2023
Master's Degree Mathematics, Vision and Learning (MVA) at ENS Paris-Saclay

2019 - 2022
French Engineering degree (=M.Sc.) in Applied Mathematics at École Centrale de Lyon

I am also passionate about Mathematics! I pursued a double-curriculum with Lyon 1 University during my engineering studies, getting a B.Sc. in Mathematics and an M.Sc in Applied Mathematics.


The Learnable Typewriter: A Generative Approach to Text Analysis
Y. Siglidis, N. Gonthier, J. Gaubil, T. Monnier, M. Aubry
ICDAR 2024
Webpage   Paper   Code

We introduce a generative approach for character analysis and recognition in text line images that can perform using little to no supervision.


Research projects

Sparse variable selection in Lasso using Knockoffs
J. Marlhens, J. Gaubil. Advisor: Prof. Y. de Castro (Institut Camille Jordan)
Winter term 2022

Study of the control of False Discovery Rate via Knockoffs as proposed by R. Barber and E. Candès, and application to System Identification.

Source : [1]
Object detection and 6D pose estimation for real-time Robotics
Advisor: Prof. E. Dellandrea (CNRS Liris)
May - Aug. 2021
Report  (french)

Literature review and evaluation of baseline Deep Learning approaches for real-time Object detection and 6D pose-estimation.

Research internship on the Robotics project Chiron in the LIRIS team, that aimed at creating a semi-autonomous robot assistant for hospitalized patients.

My contributions consisted in a 40 pages literature report of the field as well as fixing the codebase of an existing approach (FS-Net, CVPR'21) that further served as a baseline for the project. My work was then used as a basis by an incoming PhD student.

Source : [2]
Incremental Deep Learning
J. Gaubil, Clément Martinelli. Advisor: Prof. E. Dellandrea (CNRS Liris)
Academic year 2020 - 2021
Report  (french)

Study of approaches to alleviate Catastrophic Forgetting when training an already optimized Neural Network on new samples or tasks.

Academic research project during 2nd year at École Centrale de Lyon. We delivered a 50-page report over core existing Incremental Deep Learning techniques and proposed a theoretical improvement of one of them (GEM).

Course projects

Here are listed some relevant projects completed as part of course evaluations during my curriculum.

Neural Point-based rendering and View Synthesis MVA, Point Cloud and 3D Modelling course.  Report

Study of the recently proposed Scuplted Neural Points. Evaluation of the proposed sculpting mechanism, ablation study and proposition of improvements.

Self-Supervised Learning of Visual Representations MVA, Computer Vision course  with Manh-Dan Pham.  Report

Study of the recently proposed VICReg and SimCLR, evaluation of the representations learned by VICReg on a challenging downstream task of Fine-grained image Classification and in a small dataset setting.

Weakly-supervised analysis of text-line images MVA, Deep Learning course.  Report  Code

Developments of the Learnable Typewriter by using several learned prototypes for each character instead of a single one, which then made it to the final paper. Design of a richer transformation scheme by using Thin Plate Spline transformations.