Professors: Gianni Franchi (ENSTA Paris), Vicky Kalogeiton (LIX, École Polytechnique), Andrei Bursuc (Valeo.ai)
Teaching Assistants: Marwane Hariat, Firas Gabetni, Rémi Kazmierczak
Assessment: Laboratory reports and written exam
Course Description
This course provides a comprehensive introduction to modern deep learning techniques applied to computer vision tasks, with an emphasis on robustness, generalization, and interpretability. Students will learn the theoretical foundations of convolutional and generative models, explore recent advances in visual foundation models, and understand the challenges of uncertainty estimation and explainability in real-world vision systems.
Through a combination of lectures and hands-on lab sessions, students will gain practical experience with neural network architectures, segmentation, tracking, and self-supervised learning. The course culminates in a written examination and applied lab work designed to test understanding and implementation skills.
Learning Objectives
- Understand the fundamental principles of deep learning for computer vision.
- Implement and evaluate CNNs, VAEs, GANs, and diffusion models.
- Apply modern tracking and segmentation techniques in real-world scenarios.
- Analyze uncertainty, robustness, and explainability in visual models.
- Critically assess current research and trends in robust computer vision.
Evaluation
- Laboratory reports (40%)
- Written exam (60%)
- Optional bonus homeworks: +2 points per completed assignment