Fig. 11From: A survey on GAN acceleration using memory compression techniquesTraining architecture for the work [33] showing intermediate distillation loss, perceptual-loss, and gan-lossBack to article page