Research


Works at Berkeley DeepDrive

Scalabel's instance segmentation labeling tool

Scalabel: A Versatile and Scalable Annotation Tool

Core contributor for Scalabel; mainly focusing on 2D bounding box tracking, instance segmentation, and segmentation tracking annotations. Also attached various assistance models to the tool achieve human in-the-loop annotation. Accepted as a Demo in ECCV 2018.

See Project


Independent Research Projects

Predicted depth map with the KITTI dataset

Single Image Depth Estimation with Feature Pyramid Network

A simple end-to-end model with FPN that achieves state-of-the-art performance in depth prediction implemented in PyTorch.

See Code

Anicca

A Pure NumPy/CuPy Implementation of Capsule Network

NumPy, CuPy and PyTorch implementation of the paper Dynamic Routing between Capsules. Manually wrote the entire forward/backward propagation processes.

See Code


Works at Healthcare Robotics Lab

placeholder image 2

Towards Material Classification of Scenes Using Active Theomography

Pixel-wise classification of materials using thermal videos after applying a heat pulse. Designed and crafted the thermal sensing module, constructed the sensing model and trained state-of-the-art models. Accepted by IROS 2018.

Graph of distinguishable material pairs

A Model that Predicts Material Classification Performance of Thermal Tactile Sensing

Constructed a model for binary classification to predict material classification performance. Classified simulated data and real-world data collected by an 1-DOF robot to make comparison with the performance model.

See Paper