FPGA-Accelerated Convolutional Neural Network

[en] In recent years, FPGA has become an attractive solution to accelerate CNN classification for its flexibility, short time-to-market, and energy efficiency. The real-time evaluation of a CNN for image classification on a live video stream can require billions or trillions of operations per second. To come with a competitive re-configurable implementation satisfying both development time and flexibility, we propose using as a base a re-configurable Architecture composed by a set of image and video processing blocks. The whole architecture can be configured on-the-fly based on the image characteristics thus supporting variable image resolutions for each layer of the CNN.