Computational Imaging Group (CIG) is proposing the following projects for this Fall semester. If you are interested in a given project, please apply by emailing your resume to the project supervisor.

Learning for 3D computational imaging

Status: reserved (Weijie Gan) Description: Best computational imaging methods must exploit intricate relationships and features in the data. However, designing such methods for three-dimensional (3D) imaging is challenging due to the large amount of data. The goal of this project is to leverage the latest progress in machine learning to develop scalable algorithms capable of delivering highest quality results under high-data loads. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov

Convolutional dictionary learning for image reconstruction

Status: active (Fa Long) Description: Convolutional dictionary learning (CDL) is a popular model for data-driven and unsupervised image reconstruction. However, the actual process of learning a dictionary is generally computationally intractable. In this project, the goal is to leverage recent advances in optimization and regularization theory to develop algorithms for efficient CDL. The key application in this project will be magnetic resonance imaging (MRI). Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov

Unsupervised deep image learning

Status: active (Hanrui Zou) Description: One of the big challenges in applying deep learning to biomedical imaging is the lack of training data for supervised learning. The goal of this project is to overcome this limitation by looking into ways to reduce the dependence on the training data. The results of this project can be applied to large-scale imaging with limited datasets. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov

Nonconvex optimization for image restoration

Status: active (Jialong Zhang) Description: Image restoration is often formulated as an optimization problem. Convex objective functions are often more desirable due to their computational tractability. However, there is a growing interest in going beyond convexity by considering nonconvex functions that better capture the geometry of the problem. The goal of this project is to develop analytical tools and methods for nonconvex image restoration. Supervision: Ulugbek Kamilov (kamilov@wustl.edu)

Distributed algorithms for nonconvex stochastic optimization

Status: active (Melena Abijaoude) Description: Stochastic gradient descent (SGD) is one of the most widely used optimization methods for parallel and distributed processing of large datasets. To reduce the communication bottleneck of distributed SGD, recent work has considered a one-bit variant of SGD, where only the sign of each gradient element is used in optimization. The goal of this project is to extend these ideas to a stochastic variant of the proximal-gradient method that also uses one-bit per update element. Supervision: Ulugbek Kamilov (kamilov@wustl.edu)

Regularization with a convolutional neural network (CNN)

Status: active (Jiaming Liu) Description: Convolutional neural networks (CNNs) have been successfully applied to numerous tasks in computational imaging and computer vision. Currently, their application for regularization of ill-posed inverse problems is a popular topic of research. In this project, the goal is to investigate various strategies of incorporating CNNs for stabilizing image reconstruction with applications to optical imaging. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov

Learning for magnetic resonance imaging (MRI)

Status: active (Jiarui Xing) Description: Magnetic resonance imaging (MRI) is a medical imaging technique used in radiology to form pictures of the anatomy and the physiological processes of the body. One of the key practical limitations of MRI is the long duration of data acquisition. The goal of this project is to reduce the scan times by using deep-learning-based imaging techniques. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov

Learning for material imaging with light

Status: active (Ray Wu) Description: Optical imaging has been widely applied in various fields such as material imaging for nano-chip design. Recent works have shown that deep neural networks (DNNs) are able to invert multiple light scattering and to reconstruct a high-quality two-dimensional (2D) image in a very short time, which is attractive for industrial applications. The goal of this project is to design a novel DNN architecture that can reconstruct 3D objects from the light measurements and deploy the model for material imaging. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov

Image restoration with generative adversarial networks (GANs)

Status: active (Fangying Zhai) Description: Generative adversarial networks (GANs) is a promising tool for various imaging tasks. This project investigates their application to the problem of image restoration. In particular, the key challenge addressed here is their viability for restoring large images. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov