Plug and Play Network Generation PPGN is one of the neural network models, which was proposed by Nguyen et al. in 2016.

PPGN is based on approximate Langevin sampling and uses a Markov chain to generate images. The gradient of the Langevin sampler is estimated by a denoising autoencoder, which is trained using loss functions, one of which includes the GAN loss.

Features of PPGN

• Integrate adversarial training, CNN feature matching, denoising autoencoder, and Langevin sampling;
• The generated images have great differences among the same category, and can generate images of different categories and multiple categories according to the specified specifications. The generated images are clear and have high resolution.
• You can use the ImageNet 1000 classification network to generate images of a specific class.