Works at Berkeley DeepDrive

BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning

BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning

Worked as a core oraganizer of BDD100K, a dataset with 10 visual perception tasks in the context of autonomous driving. Worked with annotation vendors, compiled the datasets, and led multitask learning experiments and benchmarks. Also organized the challenges in CVPR Workshop on Autonomous Driving 2019 and 2020. Paper accepted as an Oral presentation in CVPR 2020.

See Paper

Scalabel's instance segmentation labeling tool

Scalabel: A Versatile and Scalable Annotation Tool

Worked as a core developer of Scalabel, a web-based collaborative annotation platform for various 2D and 3D tasks for visual perception aimed for democratizing and accelerating data labeling. Accepted as a demo in ECCV 2018.

See Project

Independent 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


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