Multi-class Generative Adversarial Networks with the L2 Loss Function

By Xudong Mao, Qing Li, Haoran Xie, Raymond Y. K. Lau, Zhen Wang. 2016

222 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

Citations in the corpus

A U-Net Based Discriminator for Generative Adversarial Networks (2020, background): “Plenty of efforts have been made to improve the training and performance of GANs, from reformulation of the objective function [31, 1, 26, 37], integration of different regularization techniques [51, 34, 40, 48] and architectural changes [38, 19, 13, 27].”
Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks (2017, methodology): “First, for L GAN (Equation 1), we replace the negative log likelihood objective by a least-squares loss [33].”
Feedback Adversarial Learning: Spatial Feedback for Improving Generative Adversarial Networks (2019, methodology): “We train the model to optimize the least-squares loss proposed in [38] and scale the reconstruction loss by 10.”
Speech Enhancement with a Generative Adversarial Network (2019, methodology): “7 Least Squares GAN Generative adversarial networks’ training abilities were further improved by the introduction of the least squares loss function (LSGAN; Mao et al., 2016).”
Image Manipulation with Perceptual Discriminators (2018, methodology): “Following CycleGAN, we use LSGAN [29] as an adversarial objective for that model.”
Variational Discriminator Bottleneck: Improving Imitation Learning, Inverse RL, and GANs by Constraining Information Flow (2018, background): “Alternative loss functions have been proposed to mitigate this problem (Mao et al., 2016; Zhao et al., 2016; Arjovsky et al., 2017).”
Adversarial Learning for Semi-supervised Semantic Segmentation (2018, background): “Since the GAN framework with its theoretical foundation is proposed [10], it draws significant attention with several improvements in implementation [1, 3, 8, 31, 38] and applciations including image generation [38], superresolution [22, 24], optical flow [23], object detection [42], domain adaptati”
A Generative Adversarial Approach for Zero-Shot Learning from Noisy Texts (2017, background): “Much work has been [54, 26, 23, 4, 15] proposed to improve GANs by stabilizing training behavior and eliminating mode collapse, via using alternative objective losses.”
Real-Time Monocular Depth Estimation Using Synthetic Data with Domain Adaptation via Image Style Transfer (2018, methodology): “The original work in [90] replaces the log likelihood by a least square loss to improve training stability [53].”
MSG-GAN: Multi-Scale Gradient GAN for Stable Image Synthesis (2019, result): “This is a fundamental issue with GANs, and has been widely reported by previous works [24, 19, 2, 9, 16, 28, 12, 13, 31, 22].”
PrivEdge: From Local to Distributed Private Training and Prediction (2020, background): “for increased stability during training and better quality of the reconstructed images [67].”
Image to Image Translation for Domain Adaptation (2017, background): “We found that the Wasserstein GAN is not well suited for discriminating in the Z domain since both the ‘real’ and ‘fake’ distributions are changing.”
Cross-modality image synthesis from unpaired data using CycleGAN: Effects of gradient consistency loss and training data size (2018, unknown): “(2) by least-squares loss as in [14].”
Generative Semantic Manipulation with Contrasting GAN (2017, methodology): “In practice we can replace the negative log likelihood objective by a least square loss [18], which performs more stably during training and generates higher quality results.”
One-Shot Unsupervised Cross Domain Translation (2018, methodology): “` can be the binary cross entropy or the least square loss, `(x, y) = (x− y)(2) [23].”
PA-GAN: Improving GAN Training by Progressive Augmentation (2018, methodology): “by using the Pearson χ(2) divergence (Mao et al., 2016), the Wasserstein distance (Arjovsky et al.”
One-Sided Unsupervised Domain Mapping (2017, background): “The first is to replace the negative log-likelihood by a least square loss X.Mao et al. (2016) and the second is to use a history of images for the discriminators, rather then only the last image generated Shrivastava et al. (2016).”
AGAN: Towards Automated Design of Generative Adversarial Networks (2019, background): “In LS-GAN [6], the authors consider a least square loss which corresponds to minimizing the Pearson χ2 divergence between the distributions.”
Manipulating Attributes of Natural Scenes via Hallucination (2018, background): “[2016] were the first to propose a architecture that can be trained on large scale datasets, which sparked a wave of studies aimed at improving this line of work [Arjovsky et al. 2017; Mao et al. 2016; Salimans et al. 2016].”