Image Edge Detection Base on Conditional Generative Adversarial Nets

Abstract

We propose a method to prepare for developing the quality of the reconstructing objects from edge detection. We know that some conditional generative adversarial networks like pix2pix, learn a loss function to train the mapping from input image and output image. In case of using single mapping, we cannot guarantee that all samples in X and all samples in the Y are reasonably corresponding. So, we suppose to utilize bijection and we make the optimization for the pix2pix'U-net, which can develop our model to reconstruct objects from edge detection needing to be repaired. These can let our image generated by edge detection with our method get less probability of mode collapse and ensure the image style more similar to samples.