Javad Rajabi

I am a Computer Science PhD student at the University of Toronto, co-supervised by David B. Lindell and Babak Taati, and a Faculty Affiliate Researcher at the Vector Institute. I am also a research intern at Samsung AI Center Toronto.

My research focuses on scaling diffusion and flow-matching models for image and video generation, with an emphasis on training-free methods that unlock capacity already present in pretrained models. I work along three axes: quality, designing new guidance methods that steer sampling toward better outputs; space, generating at resolutions well beyond what models were trained on while keeping compute in check; and time, accelerating the many-step sampling process so that high-quality image and video generation becomes fast and affordable.

Email CV Google Scholar Twitter GitHub LinkedIn

Research Interest

More broadly, I am interested in generative modeling with diffusion and flow models, especially in methods grounded in how these models behave internally, such as their attention patterns, frequency content, and sampling dynamics. Longer term, I am excited about generative models as a way to understand and construct rich visual worlds, from images and videos to 3D and 4D scenes. I also have a deep interest in mathematics, optimization, and dynamical systems, and in their applications to generative models and machine learning.

News

(Sep-2026) SEGA got accepted in NeurIPS 2026! 🎉
(Oct-2025) Started my research internship at Samsung AI Center Toronto!
(Sep-2025) Token Perturbation Guidance (TPG), my first paper as a PhD student got accepted in NeurIPS 2025! 🎉
(Jan-2025) Started my PhD at University of Toronto (UofT)!

Selected Publications
	AReA: Adaptive Receptive-Field Attention for Efficient High-Resolution Diffusion Transformers Javad Rajabi, Koorosh Roohi, David B. Lindell, Babak Taati Under review project page / arXiv
	SpecKCast: Spectral Kalman Forecasting for Diffusion Acceleration Javad Rajabi, Hakki Can Karaimer, David B. Lindell, Babak Taati, Michael S. Brown Under review project page / arXiv
	PhotoQuilt: Training-Free Arbitrary-Resolution Photomosaics via Bootstrapped Tiled Denoising Koorosh Roohi, Javad Rajabi**, Andrew Fleet, Babak Taati WACV 2027 project page / arXiv
	SEGA: Spectral-Energy Guided Attention for Resolution Extrapolation in Diffusion Transformers Javad Rajabi, Kimia Shaban, Koorosh Roohi, David B. Lindell, Babak Taati NeurIPS 2026 project page / arXiv
	Token Perturbation Guidance for Diffusion Models Javad Rajabi, Soroush Mehraban, Seyedmorteza Sadat, Babak Taati NeurIPS 2025 project page / arXiv
* Equal contribution

Teaching Assistant

CSC420: Introduction to Image Understanding — Winter 2025, Fall 2025, Winter 2026, Fall 2026
CSC2529: Computational Imaging — Fall 2026
CSC311: Introduction to Machine Learning — Summer 2026
CSC2515: Introduction to Machine Learning — Winter 2026
CSC263: Data Structures and Analysis — Winter 2025

Academic Service

Conference reviewer: CVPR 2026, ECCV 2026, NeurIPS 2026