CV
Saad Lahlali
Summary
PhD Candidate in Computer Vision & 3D Perception with a focus on developing novel architectures for robust multi-modal object detection and segmentation with limited labeled data using self-supervised and weakly supervised approaches.
Education
- PhD in Perception for Autonomous Driving2025-12Paris-Saclay University & CEA LISTCourses: Multi-modal object detection, Segmentation, Self-supervised learning, Weakly supervised learning
- Engineering Cycle2022Institut Polytechnique de ParisGPA: 4/4Courses: Statistics, Optimization, Object Oriented Programming, Linear Algebra, Databases, 3D Computer Vision, Deep Learning for Medical Imaging, Reinforcement Learning, Mining of Large Datasets, Learning for Robotics, Perception for autonomous systems, Explainable AI
- Mathematics and Physics2019Preparatory Classes
Work Experience
- Computer Vision Research Intern: Incremental Learning2022-04 - 2022-10CEA LISTDeveloped novel incremental learning methods for semantic segmentation in resource-constrained settings using a Transformer network, achieving state-of-the-art results.
- Research Intern: Alzheimer Detection from Handwriting2021-06 - 2021-09Institut Polytechnique de ParisDeveloped a robust algorithm using CNNs to detect early signs of Alzheimer's disease from handwriting samples. Implemented an ensemble learning technique that improved classification accuracy by 10%.
- Computer Vision Intern: Crowd Counting2020-06 - 2020-09OpinakaExplored advanced crowd counting algorithms for real-time video analytics. Designed a Scale-Attention Autoencoder with Inception Modules in TensorFlow and optimized inference time by 50% while maintaining an error rate below 25 MAE on the ShanghaiTech dataset.
- Intern: Facial Emotion Analysis2020-01 - 2020-06VocaCoachCleaned and augmented the FER2013 image dataset to enhance training data quality and reduced class imbalance through web scraping. Developed a seven-class emotion classification network using VGG19 with transfer learning.
Publications
- ALPI: Auto-Labeller with Proxy Injection for 3D Object Detection using 2D Labels Only2025
- Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion2025CVPR 2025Authors: Saad Lahlali*, Sandra Kara*, Hejer Ammar, Florian Chabot, Nicolas Granger, Hervé Le Borgne, Quoc-Cuong Pham.