Parsa Rahimi Noshanagh

Parsa Rahimi Noshanagh is a Doctoral Assistant and PhD student at the School of Engineering, École Polytechnique Fédérale de Lausanne (EPFL), affiliated with the Department of Electrical Engineering and the LIDIAP laboratory (Idiap Research Institute). He is conducting research in synthetic data generation and its applications in computer vision under the supervision of Prof. Marcel and Prof. Alahi.

• Master of Science in Electrical Engineering, Sharif University of Technology
• Currently pursuing Doctoral Program in Electrical Engineering, EPFL

His research focuses on enhancing discriminative models such as classifiers using synthetic data generated by generative models (e.g., StyleGANs, Diffusion models, Flows). He explores the concept of Analysis by Synthesis, aiming to make generative models practically useful in real-world applications beyond entertainment. His work investigates realism transfer in 3D face renderings and synthetic data augmentation to improve face recognition accuracy.

His recent publications have been accepted at top conferences including ECCVW 2024 (ORAL), ICASSP 2024, and IJCB 2023 (ORAL), demonstrating performance gains in benchmarks like IJB-C, IJB-B, and LFW. These works highlight that synthetic augmentation can rival architectural improvements in model performance.

• Paper accepted as ORAL presentation at ECCVW 2024
• Paper accepted at ICASSP 2024
• Paper accepted as ORAL presentation at IJCB 2023

Parsa previously worked as a Lead Research Engineer at MCI before starting his PhD. He has contributed to the release of datasets such as RealDigiFace and plans to release synthetic datasets from AugGen. He does not currently supervise any students but is actively involved in research collaborations with the Biometric Group at Idiap.

He is a member of the Doctoral Program in Electrical Engineering (EDEE) at EPFL and contributes to advancing privacy-conscious and resource-efficient machine learning through self-contained synthetic data generation.