Shihua Huang (黄世华)

Shihua Huang is currently a PhD student with the Dept. of Computer Science and Engineering at Michigan State University, East Lansing, USA.

His research interests are in the field of representation learning, notably neural architecture search, deep learning assisted evolutionary algorithms, and in particular dense image prediction (auto-driving and medical image analysis) and robust models.

Email  /  CV  /  简历  /  Bio  /  Google Scholar  /  Github

Given systematic ablative experiments, insights are derived for RobustScaling and RobustResblock which are then combined for RobustResNets. RobustResNets consistently outperform both the standard WRNs and other existing robust architectures, achieving state-of-the-art AutoAttack robust accuracy of 61.1% without additional data and 63.7% with 500K external data while being 2× more compact in terms of parameters.

FaPN is a simple yet effective top-down pyramidal architecture to generate multi-scale features for dense image prediction. It improves FPN's AP / mIoU by 1.5 - 2.6% on all tasks. It achieved 56.7% mIoU over ADE20k-150 when paired with MaskFormer.