Saeed Saadatnejad

PhD, Computer vision researcher

saeed2.png

{first.lastname}@epfl.ch

Office GC C1 391, EPFL,

Lausanne, Switzerland

I’m a research scientist at EPFL, where I recently earned my Ph.D. in computer science advised by Alexandre Alahi. I was also a Marie-Curie fellow as EPFLInnovator.

Research interests I’m interested in deep generative models (Normalizing Flows, VAEs, GANs, Diffusion models, …) and have used them in different applications: image synthesis, behavior prediction and human motion forecasting. I am also conducting research on analyzing failures of trajectory prediction models of self-driving cars using generative models and adversarial attacks.

Previously, I obtained my MS and BS in computer science and electrical engineering at Sharif Univ. of Tech. I have also completed research internships at Disney Research and Valeo and co-founded Pathy. Feel free to check my LinkedIn for more information. A detailed CV/resume is available upon request.

news

Jan 2024 Our paper, “Social-Transmotion: Promptable Human Trajectory Prediction” got accepted in ICLR’24.
Nov 2023 I had an awesome time visiting Bosch-AI Research in Renningen and the Karlsruhe Institute of Technology (KIT), where I gave talks on “Generic Motion Forecasting with Generative Models”. Big thanks to Andrey Rudenko, Luigi Palmieri, and Christoph Stiller for hosting me.
Nov 2023 I’ll be giving a virtual talk on Generative AI on Dec 12th at Amirkabir Univ. of Tech., AAISS event.
Oct 2023 Taking the lead at the Workshop on Visual Perception for Navigation in Human Environments at ICCV on Oct 3rd. Feel free to explore our exciting program – look forward to seeing you there!
Sep 2023 I emerged from my PhD public defense, officially bearing the title of “Doctor”! 🎓
You can access my thesis online: Deep Generative Models for Autonomous Driving: from Motion Forecasting to Realistic Image Synthesis.
Jul 2023 Launching a new challenge on end-to-end human trajectory forecasting in crowds as part of our ICCV workshop. Check it out JRDB Trajectory.
Jun 2023 I successfully passed my PhD private defense with award nominations. Special thanks to my jury members Mathieu Salzmann, Fawzi Nashashibi, Luigi Palmieri, Olga Fink and Alexandre Alahi.
May 2023 Excited to be at ICRA’23 in London! Drop me a line if you’d like to meet up and chat.
Apr 2023 Unposed paper (UNcertainty in human POSE forecasting) is out. Stay tuned for the release of our open-source library.
Mar 2023 I’m the corresponding organizer of the 4th Workshop on Visual Perception for Human Navigation, part of ICCV23 in Paris. Our impressive speaker lineup includes Marco Pavone, Patrick Perez, Fatma Guney, Georgios Pavlakos, and Ehsan Adeli. Exciting updates coming soon.
Feb 2023 The code of DePOSit, our paper on Denoising Diffusion model for human POSe forecasting in the wild, accepted in ICRA’23, was released.
Jan 2023 Yesterday, I presented a part of my research on motion forecasting to NVIDIA autonomous vehicles research lab.
Dec 2022 I have given talks on “Social-awareness and scene generalization of trajectory prediction models” at Stanford Vision and Learning (SVL) lab, UC Berkeley MSC lab and Waymo research.
Nov 2022 I’m attending NeurIPS’22 in New Orleans and will be presenting our diffusion model for pose forecasting. Feel free to send me a message if you want to chat!
Jul 2022 I’m starting a research internship at Disney research in Zurich.
Jun 2022 I’ll be attending CVPR’22 virtually!
May 2022 I’ll be giving a talk at TeIAS in few days.
Mar 2022 Scene-attack got accepted in CVPR’22: “Vehicle trajectory prediction works, but not everywhere”
Feb 2022 Social-attack got accepted in TR_C: deep analyzing the socially-awareness of trajectory prediction models
Jan 2022 For persian speakers(!), I’ll be giving a talk on “Image synthesis for simulators using generative models” at IPM.
Dec 2021 I’m giving a talk on “Generative models for self-driving cars” at Sharif Univ. of Tech., ReACT event.
Sep 2021 Our decoupled pose forecasting (DerPoF) ranked second in Somof challenge and will appear in ICCVW.
Aug 2021 Semantically-aware discriminators (SemDisc) are what we need to generate photo-realistic images with GANs. Our paper got accepted in ITS.

selected publications

  1. Social-Transmotion: Promptable Human Trajectory Prediction
    Saeed Saadatnejad, Yang Gao*, Kaouther Messaoud, and Alexandre Alahi
    In International Conference on Learning Representations (ICLR), 2023
    arXiv Code
  2. A generic diffusion-based approach for 3D human pose prediction in the wild
    Saeed Saadatnejad, Ali Rasekh, Mohammadreza Mofayezi, Yasamin Medghalchi, Sara Rajabzadeh, Taylor Mordan, and Alexandre Alahi
    In International Conference on Robotics and Automation (ICRA), 2023
    arXiv Code
  3. Vehicle trajectory prediction works, but not everywhere
    Saeed Saadatnejad, Mohammadhossein Bahari*, Ahmad Rahimi, Mohammad Shaverdikondori, Seyed-Mohsen Moosavi-Dezfooli, and Alexandre Alahi
    In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022
    arXiv Code Website
  4. Are socially-aware trajectory prediction models really socially-aware?
    Saeed Saadatnejad, Mohammadhossein Bahari*, Pedram Khorsandi, Mohammad Saneian, Seyed-Mohsen Moosavi-Dezfooli, and Alexandre Alahi
    Transportation Research Part C: Emerging Technologies (TR_C), 2022
    arXiv Code Website
  5. A Shared Representation for Photorealistic Driving Simulators
    Saeed Saadatnejad, Siyuan Li, Taylor Mordan, and Alexandre Alahi
    IEEE Transactions on Intelligent Transportation Systems (T-ITS), 2021
    arXiv Code