Baseline Methods

The baseline methods for HSLD benchmark is based on Fully convolutional networks (FCNs), SegNet, U-Net, and Feature pyramid networks (FPNs).

Backbone Networks

AlexNet

ResNet

Deep Regression Frameworks

Fully convolutional networks (FCN)

SegNet

U-Net

Feature pyramid networks (FPNs)

The Intelligent Design and Robust Learning laboratory of Defense Innovation Institute of Academy of Military Sciences focuses on the basic frontier and cross-research of artificial intelligence and aircraft design. The team has developed a series of data and model driven deep learning methods for efficient multi-disciplinary analysis and optimization of aircraft, and tries to construct a intelligent auxiliary aircraft design optimization cloud platform, namely IDaaS （Intelligent Design as a Service）. The team has already publicly released two thermal analysis benchmarks, namely the HSLD （Heat-source layout dataset） and TFRD （Temperature field reconstruction dataset）. Besides, they developed a physical informed neural network framework IDRLNet. In the past three years, they have published more than 30 related papers, and more than 30 patent applications.

Expand

Collapse

Surrogate Modeling Toolbox