Developed deep learning algorithms for vision tasks including : • Instance segmentation frameworks (CNN based and transformer based) • Graph Alignment algorithms

Research project supervised by the Sondra Lab and CentraleSupelec aiming at reconstructing high resolution images from low resolution ones from the open source UAVSAR dataset provided by NASA.• SAR Data Processing using spectral 2D filtering• Image super resolution using Vision Transformers This work led to a poster presentation at the 5th Sondra workshop held in Avignon.

Worked at the main BMW plant to develop Computer Vision algorithms based on artificial intelligence, for robots to detect 3D box poses. I worked on many topics including : • Camera Calibration (extrinsic matrix estimation, intrinsic parameters)• Motion Detection (background subtraction)• Pose detection using Human Pose estimation networks and depth maps• Creating an object detection dataset (labeling with COCOannotator)After 6 months of development and maintenance in different plants (Berlin, Leipzig, Munich), I focused on improving the already-deployed algorithms using :• Unsupervised Learning for data selection to expand the existing dataset (200 000 new images every 2 weeks)• Semi Supervised Learning to train the network on pseudo labeled data. Provided technical support and handed over a calibration package and methods to plant workers.Took part in recruiting interviews for the team.

Worked in a waste water treatment plant in Marseille during summer doing cleaning and developing in VBA excel sheets for the organisation.