publications

2023

1. WheelPose: Data Synthesis Techniques to Improve Pose Estimation Performance on Wheelchair Users

   William Huang, Sam Ghahremani, Siyou Pei, and Yang Zhang

   2023
2. Action-Conditioned Realistic Motions Generation in Sketches from 2D Trajectories

   Yipeng Zhang, Yuanyi Ding, Mingjian Lu, William Huang, Qiujing Lu, and Vwani Roychowdhury

   2023
3. Negative association between nodular stromal fibrosis and large lipid droplets in human choroid plexus epithelial cells

   Michael J. Neel, Brett A. Johnson, Jessica M. Yeung, Aarij Gora, Ashley Nguyen, Juslyn Chan, Victoria Espericueta, Sam Ghahremani, William Huang, Jonathan Zebroski, Jackson Sousa, Nathan Schacherl, Kean Ehsani, Christina M. Magana-Ramirez, Peter Chang, Daniel Gillen, and Edwin S. Monuki

   2023
4. Lipid storage by the human choroid plexus

   Brett A. Johnson, Michael J. Neel, Todd Soo, Parastou Porahang, Frances A. Goyokpin, Sara Khan, Ryan S. Salehi, Nicolas B. Maramica, William Huang, Sam Ghahremani, Peter Chang, and Edwin S. Monuki

   2023
5. Biondi body amyloid is spatially clustered in the human choroid plexus: A manual and deep learning-based study

   Michael J. Neel, Brett A. Johnson, Natarsha Vukalovich, Parastou Porahang, Chloe Zhao, William Huang, Sam Ghahremani, Andy Lin, Olivia Sandhu, Binh Nguyen, Kasra Pejhanfar, Joni L. Ricks-oddie, Peter Chang, and Edwin S. Monuki

   2023
6. Individual and Alzheimer’s disease-related differences in Biondi body amyloid morphologies in human choroid plexus epithelial cells

   Michael J. Neel, Brett A. Johnson, Todd Soo, Irene Nguyen, Natarsha Vukalovich, William Huang, Sam Ghahremani, Kimia Keshvardoot, Shuhao Chen, Zhohaer Muttalib, Seongjoon Won, Sara Khan, Michael Phelan, and Edwin S. Monuki

   2023

2023

1. Users in Wheelchairs (UIW) - A Human Centered RGB Dataset of Wheelchair Users

   William Huang, Sam Ghahremani, Siyou Pei, and Yang Zhang

   2023

   Abs Poster

   Kinematic analysis of the human body has become an integral tool in multiple fields spanning from sports performance, rehabilitation, animations, and physical modeling. While traditional marker-based solutions are commonly used in kinematic analysis, it is not suitable for “in field” tasks. This problem is further exacerbated in users with disabilities, who may face difficulties in data collection settings. Markerless video-based motion capture systems using deep learning to estimate the poses of users through camera images, offering a more accessible alternative. These systems rely on large amounts of image data of humans in different day to day actions. However, most commonly available markerless motion capture systems including MediaPipe, OpenPose, BlazePose, and Detectron2 are purely trained off able-bodied datasets like COCO and consequently perform worse on users in wheelchairs. To address this disparity, we present Users in Wheelchairs (UIW), a new dataset focused on wheelchair users. This is achieved by collecting data on common and extreme wheelchair related activities which are then placed in the context of real-world settings through manual video capture. Over 2.5k images were collected from 20 hours of video of wheelchair users in 21 different action classes from novel data collection software designed to capture and label video frames. Each instance’s keypoints were then labeled through crowd worker involvement. We provide statistical analysis on our dataset in comparison to the COCO dataset and found a 32.50 AP improvement in bounding boxes on users in wheelchairs and a 13.34 AP improvement in keypoint precision.

2022

1. Differences in Drug and Alcohol Consumption Among College Students During the COVID-19 Pandemic

   Caitlin K. Chen, Kassidy M. Ford, Leela M. Wong, Eric L. Zhao, William Huang, and Daniel Eisenberg

   2022

   Slides