With a large enough training set and some regularization strategies, D cannot just memorize the training set.Īs a result, D must learn the general rules that govern the look of images from the training set (i.e. Naturally it's G's job to generate images that trick D into believing that they are from the training set. It is D's job to take a look at an image and estimate whether it is a fake or a real image (where "real" is synonymous with "from the training set"). The forger is called G (generator) and the police D (discriminator). The basic principle of GANs is to train two networks in a kind of forger-police-relationship. Training progress of the network while learning to generate color images. The 16 selected images were the "best" ones among 1024 images according to the rating by D, hence some similarity with the training set is expected. Distance was measured by 2-Norm ( torch.dist()). (Model G32up)ġ6 generated images (each pair left) and their nearest neighbours from the training set (each pair right). (Model G32up)Ħ4 generated 32x32 grayscale cat images, rated by D as the best images among 1024 randomly generated ones. (Model G32up)ġ024 randomly generated 32x32 grayscale cat images. (Model G32up)Ħ4 generated 32x32 cat images, rated by D as the best images among 1024 randomly generated ones. (Model G32up-c) Model G32upĢ56 randomly generated 32x32 cat images. (Model G32up-c)ġ6 generated images (each pair left) and their nearest neighbours from the training set (each pair right). (Model G32up-c)Ħ4 generated 32x32 cat images, rated by D as the best images among 1024 randomly generated ones. Model G32up-c (currently default)Ģ56 randomly generated 32x32 cat images. The difference between model G32up and G32up-c is simply that G32up-c is about one layer deeper and has more convolution kernels. This model is currently the default model/architecture for G. Model G32up, grayscale: th a -colorSpace="y".This model is currently not the default for G, i.e. Model G32up, color: th a -D_iterations=2.The following images were generated by networks trained with:
NN SEQUENTIAL GRAYSCALE CODE
The script also uses code from other repositories for spatial transformers, weight initialization and LeakyReLUs. Most of the code is based on facebook's eyescream project. al., implemented as a single G (generator) as described in the blog post by Anders Boesen Lindbo Larsen and Søren Kaae Sønderby. The images are enhanced with the laplacian pyramid technique from Denton and Soumith Chintala et. This script generates new images of cats using the technique of generative adversarial networks (GAN), as described in the paper by Goodfellow et al.
0 kommentar(er)
