Private and public research in Computer Vision to advance knowledge and make the world a better place.

Competition organisationThe first one, the "Ceteris Paribus Face Challenge", consisted in creating a dataset usable to detect biases in Computer Vision models. This compettion gathered about a hundred participants competing for 2000 euros of total prize money. It brought a relevant contribution to applied and research ethics, and sensitized students to those important issues.Our second competition, still in the works, will tackle earth observation from space.

Online courses, free or not, on different AI topics. I cover various topics including Machine Learning and Deep Learning basics, dense neural networks, Deep Learning project management and data management, Convolutional Neural Networks, Generative Adversarial Networks and other forms of generative models, reinforcement learning, recurrent neural networks, and finally legal and ethical aspects of the domain. I teach different students from 16 years old upwards. For an exhaustive list, please… Show more Online courses, free or not, on different AI topics. I cover various topics including Machine Learning and Deep Learning basics, dense neural networks, Deep Learning project management and data management, Convolutional Neural Networks, Generative Adversarial Networks and other forms of generative models, reinforcement learning, recurrent neural networks, and finally legal and ethical aspects of the domain. I teach different students from 16 years old upwards. For an exhaustive list, please check out my attached portfolio. Show less

Presentations and seminars for and from students and researchers. We tackle state of the art research topics in AI. Recordings are available in the attached link.

Single image 3D reconstruction :- training and iterative improvement of a complex pipeline to reconstruct the head of a person from a selfie- cost and performance evaluation as well as optimisation in cloud environments- design and proceeding of complex testing procedures : bias evaluation, robustness evaluation, user feedback evaluation

MLOps involvement: configuration systemsDevelopment of a configuration system adapted to large scale experiments (especially in a Deep Learning context). This system makes it easier to maintain clear, qualitative and traceable experimental processes.The project is open-sourced and can be found following the attached link.

Image enhancement : explored a GAN-based idea to 'beautify' fashion images.

Semantic segmentation with little data : used data-centric AI development to design and deploy a segmentation model under the constraint of a small dataset of very correlated samples, involving complex data augmentation schemes.

Implementation of a mathematical model to control the output quality of a module in a photogrammetry pipeline.

Project involving using ODE-based neural networks to improve massively their spatial complexity

Development of a new algorithm to compress neural networks with promising perspectives.Project presented for my master thesis and successfully defended on July 24th.

Industry-driven project aimed at finding a way to compress some of the best-performing neural networks in order to run them on smaller devices, possibly using them in IoT applications.

Study concerning the efficiency of state of the art methods for image denoising, using neural networks. Research for potential advances in the subject.

Utilisation of Generative Adversarial Networks to generate databases for the training process of a classifier capable of open classification.

Study of the viability for some deep learning methods applied to face detection, face recognition and face features analysis. This work was done for the firm IntuiSense.