Open Source Summit + ELC Europe 2018

Deep Learning is the most popular and the fastest growing area in Computer Vision nowadays. OpenCV is the famous Computer Vision library. Since OpenCV 3.1 there is a deep learning module (called OpenCV DNN) introduced in the library that implements forward pass (inferencing) with deep networks, which are pre-trained using some popular deep learning frameworks, such as Caffe. tensorflow and torch. Now it support most popular networks.

This presentation will introduce OpenCV deep learning architecture, It will introduce how to setup and run a deep learning network with OpenCV DNN. It will also talk about important techniques for deep network acceleration, including convolution performance auto-tuning, layer fusion and FP16 (half-float) support. We are key contributors for this work. All these optimization are important to help user achieve best performance.