1. HarDNet-MSEG is designated for polyp segmentation. It achieved the state-of-the-art in March 2021. It remains the fastest neural network among those can achieve an accuracy level of 90.
 
 2. HarDNet-BTS is designated for brain tumor segmentation. It is one of the top eight in MICCAI Brain Tumor Segmentation Challenge (BraTS’21). The team has also been invited to give an oral presentation in the conferencewas the only Taiwan-based team getting invited."

HarDNet has successfully pushed the envelope, won global reputations,created huge economic benefits in related industry. The design principles of HarDNet is – be fully aware that a DRAM fetch is 100x time-consuming10,000x power-consuming than an on-chip arithmetic operation. Given the same accuracy constraint, HarDNet runs fastersaves more power than its counterparts (e.g., CGGResNet). HarDNet is open-sourced on GitHubthus adopted by many research institutions worldwide. So far it has received 800+ stars on GitHub.

"1. CenterNet-HarDNet has been deployed in a foundry lab for parts inspection. It can effectively minimize the misuse of defective partsthus generate few hundreds of million dollars per year.
 
 2. The HarDNet-MSEG based polyp detection technique is on its way to TFDA＇s certification process.
 
 3. The HarDNet-BTS based brain tumor detection technique is on its way to TFDA＇s certification process."