Adaptive blocked Gibbs sampling for inference in probabilistic graphical models

Islam, Mohammad Maminur; Farabi, Mohammad Khan Al; Venugopal, Deepak

doi:10.1109/ijcnn.2017.7965864

Cited by 2 publications

(4 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similar studies include multimodal image style transfer [48,49]. Inspired by coarse-to-fine method [41], Gibbs sampling [23,47,49] and attention mechanism [36,55], we propose a new progressive image generative method based on spatial recursive adversarial expansion. We first generate the central part of the image (attention part of the image), instead of generating the whole image in once, and then gradually recursively generate the surrounding image content based on previously generated content.…”

Section: Related Workmentioning

confidence: 99%

“…In terms of the number of samples in the current dataset, the learned distribution can not represent the actual spatial distribution of high-resolution images, resulting in poor generated image quality and other problems. In order to solve this problem, inspired by the collapsed Gibbs sampling, [23,47,48] we iteratively generate the whole image from the central part of the image instead of generating the whole one at once, which greatly reduces the spatial dimension of the image we need to learn and generate. So we propose a solution to divide an image into several different non-overlapping image blocks.There are several ways to divide an image into different image patches, inspired by attention mechanism, we will divide the image from the image center.…”

Section: Overall Architecture and Learning Schemesmentioning

confidence: 99%

“…. ., N.A basic attention network (BAAN) can be one of the typical GAN, such as DCGAN [1], BEGAN [47], WGAN-GP [23], which mainly responsible for learning the distribution of attention patches. All spatial adversarial expansion networks will be cascaded together recursively.…”

Section: Overall Architecture and Learning Schemesmentioning

confidence: 99%

“…Therefore, the network that generates high-resolution images is prone to unstable training or even unable to converge, and it is also prone to generate many non-sense images. In order to solve the problem of quality and stability in high-resolution image generation, inspired by Collapsed Gibbs sampling [23,47,48] and attention mechanism , this paper proposes a spatial recursive generative adversarial network (SpatialGAN). The SpatialGAN first generates the central part of the image, whose spatial dimension (spatial resolution) is relatively small, the variability of the central part content is greatly reduced, which enhances the reliability of the network.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

SpatialGAN

Zhao

Lin

et al. 2020

Proceedings of the 28th ACM International Conference on Multimedia

View full text Add to dashboard Cite

The image generation model based on generative adversarial networks has recently received significant attention and can produce diverse, sharp, and realistic images. However, generating high-resolution images has long been a challenge. In this paper, we propose a progressive spatial recursive adversarial expansion model(called SpatialGAN) capable of producing high-quality samples of the natural image. Our approach uses a cascade of convolutional networks to progressively generate images in a part-to-whole fashion. At each level of spatial expansion, a separate image-toimage spatial adversarial expansion network (conditional GAN) is recursively trained based on context image generated by previous GAN or CGAN. Unlike other coarse-to-fine generative methods that constraint on generative process either by multi-scale resolution or by hierarchical feature, the SpatialGAN decomposes image space into multiple subspaces and gradually resolves uncertainties in the local-to-whole generative process. The SpatialGAN greatly stabilizes and speeds up the training, which allows us to produce images of high quality. Based on visual Inception Score and Fréchet Inception Distance, we demonstrate that the quality of images generated by SpatialGAN on several typical datasets is better than that of images generated by GANs without cascading and comparative with the state of art methods with cascading.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Overall Architecture and Learning Schemesmentioning

confidence: 99%

Section: Overall Architecture and Learning Schemesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations