DeLiGAN: Generative Adversarial Networks for Diverse and Limited Data

By Swaminathan Gurumurthy, Ravi Kiran Sarvadevabhatla, R. Venkatesh Babu. 2017

305 citations recorded by the source corpus.

A class of recent approaches for generating images, called Generative Adversarial Networks (GAN), have been used to generate impressively realistic images of objects, bedrooms, handwritten digits and a variety of other image modalities. However, typical GAN-based approaches require large amounts of training data to capture the diversity across the image modality. In this paper, we propose DeLiGAN – a novel GAN-based architecture for diverse and limited training data scenarios. In our approach, we reparameterize the latent generative space as a mixture model and learn the mixture models parameters along with those of GAN. This seemingly simple modification to the GAN framework is surprisingly effective and results in models which enable diversity in generated samples although trained with limited data. In our work, we show that DeLiGAN can generate images of handwritten digits, objects and hand-drawn sketches, all using limited amounts of data. To quantitatively characterize intra-class diversity of generated samples, we also introduce a modified version of inception-score, a measure which has been found to correlate well with human assessment of generated samples.

Subfields: Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

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

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

Training Generative Adversarial Networks with Limited Data (2020, methodology): “We also try modifying our baseline to use a shallower mapping network, which can be trained with less data, borrowing intuition from DeLiGAN [16].”
Large Scale Image Completion via Co-Modulated Generative Adversarial Networks (2021, background): “Great research interest has been drawn on the evaluation of GANs (DeVries et al., 2019; Gurumurthy et al., 2017; Sajjadi et al., 2018; Snell et al., 2017; Xiang & Li, 2017).”
DiffWave: A Versatile Diffusion Model for Audio Synthesis (2020, background): “• Modified Inception Score (mIS) (Gurumurthy et al., 2017) measures the within-class diversity of samples in addition to IS.”
Adversarial AutoAugment (2019, neutral): “As shown in Figure 6(a) and 6(b),we further visualize the probability distribution of the parameters of the augmentation policies learned with PyramidNet+ShakeDrop on CIFAR-10 over time.”
It's Raw! Audio Generation with State-Space Models (2022, background): “• Modiﬁed Inception Score (mIS) [14] provides a measure of both intra-class in addition to inter-class diversity.”
The Effectiveness of Data Augmentation in Image Classification using Deep Learning (2017, background): “Furthermore, GANs have been effective even with relatively small sets of data [7] by performing transfer learning techniques.”
Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs (2020, background): “…task to facilitate training [7, 26, 48, 49], improving an inference procedure of suboptimal models [2, 39, 29, 38], using an expressive prior distribution [13], actively choosing samples to give supervision for conditional generation [29], or actively sampling mini-batches for training [37].”
ResizeMix: Mixing Data with Preserved Object Information and True Labels (2020, background): “Integrated with adversarial training [2, 21, 40], AdvAA [54] makes networks learn more hard data samples, in which the domain of dataset becomes more widespread.”
On the Evaluation of Conditional GANs (2019, background): “Most of these metrics either focus on the separability between generated images and real images [21, 30, 50, 16], compute the distance between distributions [10, 12, 2], assess sample quality and diversity from conditional or marginal distributions [33, 11, 55], measure the similarity between genera”
ClusterGAN : Latent Space Clustering in Generative Adversarial Networks (2018, background): “In [11], the authors sampled from a Gaussian mixture prior and obtained diverse samples even in limited data regimes.”
Competitive Training of Mixtures of Independent Deep Generative Models (2018, unknown): “, 2016] and GANs [Gurumurthy et al., 2017].”
Data Augmentation techniques in time series domain: a survey and taxonomy (2022, methodology): “In these cases, traditional GAN metrics, such as the Inception Score[88], Mode Score[36] or Fréchet Inception Distance[40] are used to evaluate the results.”
Unsupervised multi-modal Styled Content Generation (2020, methodology): “Several works, such as [Ben-Yosef and Weinshall 2018; Gurumurthy et al. 2017; Pandeva and Schubert 2019], cope with the multi-modal case by structuring the latent spaceZ as a Mixture of Gaussians, whose first and secondmoments are learned.”
BAGAN: Data Augmentation with Balancing GAN (2018, background): “In general, it is difficult to train a GAN starting from a very little dataset, the GAN must have many examples to learn from [16].”
Generative Adversarial Networks (GANs) (2020, methodology): “-BiGAN [78], HALI [79], AGE [80], VEEGAN [81], MGGAN [82] Improved Discriminator EBGAN [83], BEGAN [84], MAGAN [85], Max-Boost-GAN [86] Memory Networks MemoryGAN [44] Latent space engineering DeLiGAN [45], NEMGAN [87], MultiplicativeNoise [88], DE-GAN [89], InfoGAN [90] New Loss Function New probabi”
Partition-Guided GANs (2021, methodology): “To obtain a disconnected latent space, DeLiGAN [22] samples z from a mixture of Gaussian, while Odena et al. [59] add a discreet dimension to the latent variable.”
Generative Adversarial Networks: An Overview (2017, methodology): “Similar to using an encoding process to model the distribution of latent samples, Gurumurthy et al. [35] propose modelling the latent space as a mixture of Gaussians and learning the mixture components that maximize the likelihood of generated data samples under the data generating distribution.”
An Improved Evaluation Framework for Generative Adversarial Networks (2018, background): “Researchers have been continuously developing better GAN architectures [18,19] and training strategies [20,21] on generating domainspecific images.”