GTC Silicon Valley-2019: Transfer Learning-Based GPU-Accelerated Deep Learning for End-to-End Industrial Inspection
Note: This video may require joining the NVIDIA Developer Program or login
GTC Silicon Valley-2019 ID:S9291:Transfer Learning-Based GPU-Accelerated Deep Learning for End-to-End Industrial Inspection
Andrew Liu(NVIDIA),Peter Pyun(NVIDIA)
We'll explain how to deploy a weakly supervised deep learning accelerator solution for end-to-end industrial inspection in the framework of vision-based anomaly detection. Compared with other computer vision problems, deploying a deep learning-based industrial inspection solution in production runs into three fundamental limiters scarcity of curated data, extreme low quantity of defects, and multi-scale defect sizes. We will describe how to overcome these by using weakly supervised transfer learning that's based on award-winning U-net architecture. We also provide a deep dive into concepts by using examples that include data curation, verification, and end-to-end deployment based on an NVIDIA GPU Cloud development environment with the latest versions of CUDA and cuDNN, as well as GPU-Accelerated Tensorflow and Keras.