1. [Publications](/index.php/publications)
2. Automated Synthetic-to-Real Generalization
 
 # Automated Synthetic-to-Real Generalization

  ![](/sites/default/files/styles/wide/public/publications/teaser2.png?itok=MO36nQEG)

 Models trained on synthetic images often face degraded generalization to real data. As a convention, these models are often initialized with ImageNet pre-trained representation. Yet the role of ImageNet knowledge is seldom discussed despite common practices that leverage this knowledge to maintain the generalization ability. An example is the careful hand-tuning of early stopping and layer-wise learning rates, which is shown to improve synthetic-to-real generalization but is also laborious and heuristic. In this work, we explicitly encourage the synthetically trained model to maintain similar representations with the ImageNet pre-trained model, and propose a *learning-to-optimize* (L2O) strategy to automate the selection of layer-wise learning rates. We demonstrate that the proposed framework can significantly improve the synthetic-to-real generalization performance without seeing and training on real data, while also benefiting downstream tasks such as domain adaptation. Code is available at: <https://github.com/NVlabs/ASG>.



 ## Authors



Wuyang Chen (Texas A&amp;M)

[Zhiding Yu](/index.php/person/zhiding-yu)

Zhangyang Wang (Texas A&amp;M)

Anima Anandkumar (NVIDIA)

 

 

 ## Publication Date



Tuesday, July 14, 2020

 

 ## Published in



[International Conference on Machine Learning (ICML) 2020](https://icml.cc/virtual/2020)

 

 ## Research Area



[Computer Vision](/index.php/research-area/computer-vision)

[Robotics](/index.php/research-area/robotics)

 

 

 ## External Links



[Project Page](https://sites.google.com/nvidia.com/asg)

[Paper (arXiv)](https://arxiv.org/abs/2007.06965)

[Code](https://github.com/NVlabs/ASG)

[Slides](https://chrisding.github.io/publications/ICML20b_Slides.pdf)