High-Fidelity Image Generation With Fewer Labels

Abstract

Deep generative models are becoming a cornerstone of modern machine learning. Recent work on conditional generative adversarial networks has shown that learning complex, high-dimensional distributions over natural images is within reach. While the latest models are able to generate high-fidelity, diverse natural images at high resolution, they rely on a vast quantity of labeled data. In this work we demonstrate how one can benefit from recent work on self- and semi-supervised learning to outperform the state of the art on both unsupervised ImageNet synthesis, as well as in the conditional setting. In particular, the proposed approach is able to match the sample quality (as measured by FID) of the current state-of-the-art conditional model BigGAN on ImageNet using only 10% of the labels and outperform it using 20% of the labels.

arXiv commentMario Lucic, Michael Tschannen, and Marvin Ritter contributed equally to this work. ICML 2019 camera-ready version. Code available at https://github.com/google/compare_gan

References in corpus (14)

• GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Self-Attention Generative Adversarial Networks
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• Conditional Image Synthesis With Auxiliary Classifier GANs
• Pixel Recurrent Neural Networks
• A Learned Representation For Artistic Style
• Semi-Supervised Learning with Generative Adversarial Networks
• A Note on the Inception Score
• Assessing Generative Models via Precision and Recall
• Improvements to context based self-supervised learning
• Grasp2Vec: Learning Object Representations from Self-Supervised Grasping
• Generating High Fidelity Images with Subscale Pixel Networks and Multidimensional Upscaling
• Semi-supervised Conditional GANs

Cited by in corpus (33)

• Diffusion Models Beat GANs on Image Synthesis
• Large Scale Adversarial Representation Learning
• Self-supervised learning methods and applications in medical imaging analysis: A survey
• Selective Synthetic Augmentation with HistoGAN for Improved Histopathology Image Classification
• Image Augmentations for GAN Training
• Differentiable Augmentation for Data-Efficient GAN Training
• StarGAN v2: Diverse Image Synthesis for Multiple Domains
• Unsupervised Object Segmentation by Redrawing
• Is Attention Better Than Matrix Decomposition?
• A Large-Scale Study on Regularization and Normalization in GANs
• A Self Supervised StyleGAN for Image Annotation and Classification with Extremely Limited Labels
• Classification and reconstruction of optical quantum states with deep neural networks
• Text to Image Synthesis using Stacked Conditional Variational Autoencoders and Conditional Generative Adversarial Networks
• cGANs with Multi-Hinge Loss
• Vision and Language: from Visual Perception to Content Creation
• Exploring Self-Supervised Regularization for Supervised and Semi-Supervised Learning
• Revisiting Image Aesthetic Assessment via Self-Supervised Feature Learning
• Data-Efficient GAN Training Beyond (Just) Augmentations: A Lottery Ticket Perspective
• Image Enhanced Rotation Prediction for Self-Supervised Learning
• What If We Only Use Real Datasets for Scene Text Recognition? Toward Scene Text Recognition With Fewer Labels
• Countering Noisy Labels By Learning From Auxiliary Clean Labels
• MatchGAN: A Self-Supervised Semi-Supervised Conditional Generative Adversarial Network
• Attentive Normalization for Conditional Image Generation
• Semi-supervised Learning for Few-shot Image-to-Image Translation
• Semi-supervised Learning using Adversarial Training with Good and Bad Samples
• Double cycle-consistent generative adversarial network for unsupervised conditional generation
• Diverse Image Generation via Self-Conditioned GANs
• Omni-GAN: On the Secrets of cGANs and Beyond
• Robust conditional GANs under missing or uncertain labels
• Weakly Supervised Context Encoder using DICOM metadata in Ultrasound Imaging
• Noise Robust Generative Adversarial Networks
• Class Balancing GAN with a Classifier in the Loop
• Instance-Conditioned GAN