ECE 697 final projects, Professor Malloy
Thursday August 6th 2020 (10AM – 12:20PM)
10:00 -10:20: Mahalakshmi Sundaresan
Automated MRI Prescription of livers. This project studies localizing and drawing a bounding box over the region containing the liver in MRI images, eliminating the need for manual intervention to perform this task. The project utilizes the YOLOv3 algorithm, and could help emerging rapid imaging protocols such as the fat and iron quantification protocol developed here at UW. [codebase]
10:30 – 10:50: Pratiksha Pai
Usage of Neural Activity Pattern Generated Cochleagram for Tempo Estimation with a Convolutional Neural Network. This project uses a Cochleagram trained Convolutional Neural Network (CNN) model to improve tempo estimation, an important part of Signal Processing Multimedia applications; the current best approach is the Mel Spectrogram with classification-learning based CNN. This Cochleagram-trained CNN is derived from Neural Activity Patterns and is based on Lyon’s Auditory Model. The resulting CNN has 3 convolutional layers, 4 multi-filter modules, and 4 dense layers. [codebase]
11:00 -11:20: Zeyang Fan
Watermark Image Alignment Based on Machine Learning. The project’s goal is to align images of handwritings, some of them would include watermarks. The image alignment part is done by neural networks instead of traditional methods. [codebase]
11:30 – 11:50: Hamsalekha Premkumar
Predicting Overall Survival of Patients Diagnosed with Glioma. Glioma is a type of tumor that occurs in the nervous system and is a cause significant morbidity and mortality. Prognostic information aids treatment and patient management. The aim of this project is to use Deep Learning to extract features from MRI sequences that can predict patients’ overall survival. [codebase]
12:00 – 12:20 PM: Haley Massa
Is There a Best U-Net Architecture? Since the introduction of the U-Net in 2015, it has been widely used for semantic segmentation tasks within medical imaging. However, every research team makes minor adjustments to the U-Net architecture. This project uses neural architecture search techniques to explore the architectural parameter space of the U-Net on different medical imaging datasets to determine if there are parameters that are consistent with a best U-Net architecture. [codebase]
Friday August 7th 2020 (10AM – 12:20PM)
10:00 -10:20: Sreeranjani Didugu
Empirical Study of Deep Convolutional GAN and Cycle GAN. Deep convolutional GANs are used to generate large number of fake images by training on a huge training dataset images e.g. human face images, MNIST dataset, etc. Cycle GANs are used for image to image translation of unpaired images, e.g. the popular horse to zebra (GIF) image translation. This project explores understanding and implementation of DCGAN and Cycle GAN algorithms. [codebase]
10:30 – 10:50: Yang Fang
Personalized Data Generation using GAN. Personalized data generation is to fill missing data in the dataset with customized requirements. This project dives into the application of Generative Adversarial Networks(GAN) to solve the nonlinear matrix completion problem. [codebase]
11:00 -11:40: Nawal Dua, Neel Kelkar, Suchith Suresh
Dynamic Depth of Field with Eye Tracking. Refocusable cinematic Depth of Field in videos is simulated using their depth maps by applying gaussian blurs to regions of a frame that are not areas of focus for the viewer’s eye. [codebase]
12:00 – 12:20: Ananya Jegannathan
Book Recommender System. This project aims to build a book recommendation system using user-book ratings data taken from Goodreads. This is executed with the help of matrix completion using iterative singular value thresholding and clustering algorithms. [codebase]