Tzer-Jen Wei
College of Artificial Intelligence, National Chiao Tung University
Deep Convolutional Neural Network has shown great success in lots of tasks in computer vision and image processing, including classification, detection, segmentation, generation, translation and more. Convolutional networks typically work on pixels and images are treated as bitmaps. We propose an architecture of convolutional networks which by design corresponds to curves and enclosed regions on spatial domain. By rendering bitmaps, the outputs can be fed into convolutional neural models, and incorporated with other superivesd, semi-supervised, or self supervised frameworks, curve and regions features can be extracted. Applications include vector graphics generation, super resolution and image quantization.
Keywords：Deep Learning, Computer Vision