Our research focuses at the intersection of computer vision, machine learning and computer graphics, to enable computers to reliably perceive our world through images.

We draw inspiration from the fact that visual scenes in digital images are naturally composed of a hierarchy of entities (e.g. objects, parts, etc.) that interact with each other in the 3D world. However, current visual recognition systems process images mostly in 2D only, which makes them brittle and unreliable. Our working hypothesis is that vision systems need to develop a causal 3D understanding images by following an analysis-by-synthesis approach. Our work demonstrates that such generative vision models are robust, learn efficiently and can solve many vision tasks at once, making them a truly foundational component of AI systems.

Our research on generative vision is a highly collaborative endeavour. We have very close long-term collaborations with the labs of Prof. Alan Yuille (JHU, US) and Prof. Christian Theobalt (MPII, Germany), and are always looking forward for exciting new projects.

We are constantly looking for highly motivated PhD students, visiting students and interns to join our team.

News

We organize 3 Workshops at ECCV 2022
April 2022


We cover different aspects of robustness in computer vision:
1) Out-of-distribution generalization
2) Cross-dataset generalization
3) Adversarial Robustness

4 Papers (1 Oral) at CVPR 2022!
March 2022


For more details, PDFs and code check out our publications.

Research Group at MPII
February 2022


Adam Kortylewski was appointed as research group leader at the Max Planck Instutite for Informatics! He will lead the Generative Vision research group with a focus on developing an analysis-by-synthesis approach to computer vision.

Emmy Noether Grant!
December 2021


Adam Kortylewski was awarded an Emmy Noether grant. He will receive 1.6 Million Euros to research how deep learning and generative models can be combined to better understand the 3D compositional structure of visual scenes in images.

Key Publications

Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose
Angtian Wang, Shenxiao Mei, Alan Yuille, Adam Kortylewski
Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose
in Advances in Neural Information Processing Systems (NeurIPS), 2021.
NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation
Angtian Wang, Adam Kortylewski, Alan Yuille
NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation
in International Conference on Learning Representations (ICLR), 2021.
Compositional Convolutional Neural Networks: A Robust and Interpretable Model for Object Recognition under Occlusion
Adam Kortylewski, Qing Liu, Angtian Wang, Yihong Sun, Alan Yuille
Compositional Convolutional Neural Networks: A Robust and Interpretable Model for Object Recognition under Occlusion
in International Journal of Computer Vision (IJCV), 2020.