Azade Farshad
Assistant Professor, Aalto University ยท Principal Investigator, ELLIS Institute Finland
My group develops generative models and structured scene representations to understand, simulate, and augment the visual world โ from natural scenes to the operating room.
About
I am an Assistant Professor at the Department of Computer Science, Aalto University, and a Principal Investigator at the ELLIS Institute Finland. Before moving to Finland in 2025, I was a postdoctoral research associate at the Technical University of Munich (TUM), where I also completed my PhD summa cum laude in 2023 with a thesis on learning neural representations with limited data and supervision โ recognized with the BVM 2025 Best Dissertation Award.
My research develops deep generative models โ with a focus on diffusion models and structured representations such as scene graphs โ to understand, generate, and manipulate images and videos. A central goal of my group is to bring these methods into the operating room: simulating surgery, modeling surgical workflow, and synthesizing realistic medical data to build the next generation of clinical AI tools. I continue to co-advise doctoral researchers at TUM with Prof. Nassir Navab.
Beyond research, I serve the community as an Area Chair for MICCAI, an Executive Committee member of the British Machine Vision Association (BMVA), a Board Member of Women in MICCAI, and a Lead Member of Women in AI and Robotics. I have organized workshops at CVPR, ICCV, and ECCV, and I am committed to mentoring junior researchers and supporting women in AI. See my full CV โ
Research
Generative Models for Medical Imaging
We build diffusion-based methods for synthesizing and manipulating medical images and videos โ counterfactual generation, physics-informed multimodal MRI synthesis, and anatomy-aware inpainting for clinically meaningful data augmentation.
Surgical Data Science & Simulation
We simulate surgery with generative video models, surgical scene graphs, and workflow understanding โ towards training tools and intraoperative assistance for the operating room of the future.
Scene Graphs & Semantic Scene Understanding
We study structured representations of visual scenes: scene-graph-guided image synthesis and manipulation, unconditional scene graph generation, and disentangled semantic editing.
Learning with Limited Data
We make medical AI work where labels and data sharing are scarce โ meta-learning, few-shot and self-supervised segmentation, and federated learning for non-IID clinical data.
My Group full team & collaborators โ
I work with doctoral researchers at Aalto University and โ jointly with Prof. Nassir Navab โ at TU Munich.
News
- 2026 ๐ We are recruiting! Open PhD and postdoc positions in generative models and medical AI.
- 2026 ๐ฃ I am co-organizing the MELEX workshop (More Exploration, Less Exploitation) at ECCV 2026.
- 2026 ๐ Conformable Convolution, on topologically constrained learning of complex anatomical structures, was accepted at AAAI 2026.
- Sep 2025 ๐ MoViS won the Best Paper Award at the CREATE Workshop, MICCAI 2025.
- 2025 ๐ Two papers accepted at MICCAI 2025: HieraSurg, our hierarchy-aware diffusion model for surgical video generation, and DeepAF, a one-shot spatiospectral auto-focus model for digital pathology.
- 2025 ๐ซ๐ฎ I joined Aalto University as an Assistant Professor and the ELLIS Institute Finland as a Principal Investigator.
- Mar 2025 ๐ My doctoral thesis received the BVM 2025 Best Dissertation Award.
- 2025 ๐ Latent Drifting, on counterfactual medical image synthesis with diffusion models, was accepted at CVPR 2025.
- Oct 2024 ๐ Two Best Paper Awards at MICCAI 2024 โ GRAIL Workshop (SANGRIA) and EARTH Workshop (VISAGE).
Selected Publications full list โ
-
Conformable Convolution for Topologically Constrained Learning of Complex Anatomical Structures
AAAI 2026 AAAI
-
HieraSurg: Hierarchy-Aware Diffusion Model for Surgical Video Generation
MICCAI 2025 MICCAI
-
Latent Drifting in Diffusion Models for Counterfactual Medical Image Synthesis
CVPR 2025 CVPR
-
MoViS: Motion-Guided Video Generation for Laparoscopic Surgery
CREATE Workshop @ MICCAI 2025 ยท ๐ Best Paper Award MICCAI
-
Surgical Flow Masked Autoencoder for Event Recognition
Medical Imaging with Deep Learning (MIDL) 2025 MIDL
-
VISAGE: Video Synthesis Using Action Graphs for Surgery
EARTH Workshop @ MICCAI 2024 ยท ๐ Best Paper Award MICCAI
-
SceneGenie: Scene Graph Guided Diffusion Models for Image Synthesis
SG2RL Workshop @ ICCV 2023 ยท ๐ Best Paper Award ICCV
-
Y-Net: A Spatiospectral Dual-Encoder Network for Medical Image Segmentation
MICCAI 2022 MICCAI
-
Semantic Image Manipulation Using Scene Graphs
CVPR 2020 CVPR
* denotes equal contribution.