Wenzhen is a computer science and math double-major student at Washington University in St. Louis. She just finished her undergraduate this spring semester and is currently working in the Computer Graphics Lab at WUSTL and doing research in geometry processing and shape understanding. She might graduate in 2018 with a master’s degree in computer science with a machine learning specialized certification, or she will do a one-year co-op internship and graduate in 2019. Her research interests are primarily in machine learning, computer vision and computer graphics. She is also broadly interested in robotics, healthcare technology, AI strategy game playing, etc. Currently, her goal is to attain a PhD in computer science and to utilize machine learning and computer vision to build smart healthcare applications. She also has an unrealistic dream—to solve the anti-aging problem, which she doesn’t have any clue about yet, but it is a direction she is hoping to go in the future.
Segmentation Using Mathematical Morphology »
Project: Instance Segmentation
Goal of the project:
Performing instance segmentation, which is to detect objects in an image while simultaneously generating a high-quality segmentation mask for each instance.
Summary of work:
I initially wanted to implement a Mask R-CNN evaluator, but the TensorFlow graph was too complicated, so we borrowed ideas from the “Mask R-CNN” paper, which is to do a region proposal first, crop the bounding box and train the region of the object with corresponding masks. The “Mask R-CNN” paper originally used FCN to perform region-to-mask training, but we used ENet, which is lighter and easier to train.
Results and future work:
Results are promising. Future work: 1. I will keep working on building the Mask R-CNN architecture. 2. I want to make an iOS app that can take a picture and then crop the objects and where the user can change the background.