Convolutional neural networks (CNNs) are a special type of feed-forward networks inspired by the visual system. They are the state-of-the-art approach for many pattern recognition tasks. Caffe is a gpu-optimized deep-learning framework, developed by the Berkeley Vision and Learning Center (BVLC). In this talk we will give an overview of CNNs in combination with Caffe in the context of object recognition. Our results cover the visualization of the network, the analysis of the architecture and the stability of the recognition. Moreover, benchmarks on existing and own datasets will be shown. Finally, a short demonstration of the capabilities of our system will be given.
Ort: INB Seminarraum
