|
Senior Machine Learning Scientist at Prescient Design (Genentech) Email / Resume / Bio / Google Scholar / LinkedIn I am a Senior Machine Learning Scientist at Genentech on the Prescient Design team. I work on ML research for molecular drug design with a focus on oncology. Prior to joining Prescient, I was a Research Scientist at Meta Reality Labs where I worked on generative models, including diffusion models and variational autoencoders for 3D objects and content creation for AR & VR. Before coming to Meta, I was a Postdoctoral Research Fellow at Johns Hopkins University working with Prof. Rama Chellappa on geo-localizing natural images. I received my Ph.D. from Rice University in 2021 where I was fortunate to work with Prof. Ashok Veeraraghavan in the Computational Imaging Group. My Ph.D. research focused on deep learning, computer vision, and computational imaging for robust unconstrained camera-based vital signs monitoring, known as imaging photoplethysmography. |
|
|
|
|
Ewa M. Nowara, Daniel McDuff, Ashok Veeraraghavan Communications of the ACM, 2022 In this article, we review state-of-the-art physiological and medical imaging modalities that leverage recent advances in computational photography. We explain the principles behind them, and discuss their advantages and their limitations, with the hope of introducing this emerging field and research avenues to a broader audience. |
|
Shraman Pramanick, Ewa M. Nowara, Josh Gleason, Carlos D. Castillo, Rama Chellappa ECCV, 2022 arXiv / poster / video We propose a novel approach to geolocalize an image taken anywhere in the world in any unconstrained settings. Our model, called TransLocator, is based on a unified dual-branch Vision Transformer (ViT) model which uses features from an RGB image and its semantic segmentaion representation. The ViT architecture allows it to attend to tiny details over the entire image and the semantic segmentaion represenation gives it robustess even under extreme appearance variations. |
|
Ewa M. Nowara, Daniel McDuff, Ashok Veeraraghavan ICCV, 2021 arXiv / video We exploit the idea that statistics of corruptions may be shared between the video regions that contain the signal of interest and those that do not. We use the inverse of an attention mask to generate a corruption estimate that is then used to denoise temporal observations. |
 
|
Ewa M. Nowara, Daniel McDuff, Ashok Veeraraghavan CVPR Workshops, 2021 We improve the generalizability of deep learning models for non-contact cardiac monitoring by augmenting the training set videos with ``magnified'' videos. These data augmentations are specifically geared towards revealing useful features for recovering the physiological signals. |
|
Daniel McDuff, Ewa M. Nowara CHI, 2021 arXiv / video We animate blood flow patterns to augment the appearance of synthetic avatars and photo-realistic faces based on a data-driven physiological model. |
|
Ewa M. Nowara, Daniel McDuff, Ashok Veeraraghavan Biomedical Optics Express, 2021 video We show that deep learning models can learn how noise at different video compression levels affects the physiological signals and are able to reliably recover vital signs from highly compressed videos, even in presence of large motion. |
|
Ewa M. Nowara, Tim K. Marks, Hassan Mansour, Ashok Veeraraghavan Trans. on Intelligent Transportation Systems, 2020 video We demonstrate that we can reduce most outside light variations using narrow-band near-infrared (NIR) video recordings and obtain reliable heart rate estimates. We present a novel optimization algorithm, which we call AutoSparsePPG, that leverages the quasi-periodicity of physiological signals and achieves better performance than the state-of-the-art methods. |
|
Ewa M. Nowara, Daniel McDuff, Ashok Veeraraghavan CVPR Workshop, 2020 video We evaluate how much gender and skin tone affect vital signs estimation from video. We find that the performance drops significantly on videos of people with very dark skin tones, especially for machine learning algorithms. |
|
Genki Nagamatsu, Ewa M. Nowara, Amruta Pai, Ashok Veeraraghavan Hiroshi Kawasaki EMBC, 2020 We use 3D face tracking to estimate the position of facial landmarks with pixel-level accuracy to improve motion robustness of camera-based vital sign estimation. |
|
Ewa M. Nowara, Daniel McDuff ICCV, 2019 We demonstrate that very small intensity variations in the skin related to physiological signals can be recovered even from very compressed videos with supervised deep learning. |
|
Ewa M. Nowara, Tim K. Marks, Hassan Mansour, Ashok Veeraraghavan CVPR Workshops, 2018 We demonstrate the feasibility of using narrow-bandwidth near-infrared (NIR) active illumination at 940 nm for camera-based vital signs measurements. We develop a novel signal tracking and denoising algorithm (SparsePPG) based on Robust Principal Components Analysis and sparse frequency spectrum estimation. |
|
Ewa M. Nowara, Ashutosh Sabharwal, Ashok Veeraraghavan Face and Gesture, 2017 We developed a machine learning system to prevent face spoofing attacks by detecting and analyzing the heartbeat signal from the face videos. |
|
T Marks, H Mansour, E Nowara, Y Nakamura, A Veeraraghavan
T Marks, H Mansour, E Nowara, Y Nakamura, A Veeraraghavan
|
|
|
|
Ewa M. Nowara Electrical and Computer Engineering, Rice University, April 2021 |
|
|
|