Sliced Wasserstein Generative Models

Wu, Jiqing; Huang, Zhiwu; Acharya, Dinesh U; Li, Wen; Thoma, Janine; Paudel, Danda Pani; Gool, Luc Van

doi:10.1109/cvpr.2019.00383

Cited by 87 publications

(77 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3.1 the sample complexity of the Wasserstein and sliced Wasserstein distances. We show that for a certain class of distributions the Wasserstein distance has an exponential sample complexity, while the sliced Wasserstein distance [8,34] has a polynomial sample complexity.…”

Section: Introductionmentioning

confidence: 99%

“…We first apply this intuition to analyze the recently proposed sliced Wasserstein distance GAN, which is based on the average Wasserstein distance of the projected versions of two distributions along a few randomly picked directions [8,20,34]. We prove that the sliced Wasserstein distance is generalizable for Gaussian distributions (i.e., it has polynomial sample complexity), while Wasserstein distance is not, thus partially explaining why [8,20,34] may exhibit better behavior than the Wasserstein distance [2].…”

Section: Introductionmentioning

confidence: 99%

“…The cause behind this was noted in [33], where it was shown that estimates of the Wasserstein distance suffer from the 'curse of dimensionality.' To tackle the instability and complexity, a sliced version of the Wasserstein-2 distance was employed by [8,20,18,34], which only requires estimating distances of 1-d distributions and is, therefore, more efficient. The "sliced Wasserstein-p distance" [5] between distributions P d and P g is defined as…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Max-Sliced Wasserstein Distance and Its Use for GANs

Deshpande

Sun

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

104

View full text Add to dashboard Cite

Generative adversarial nets (GANs) and variational auto-encoders have significantly improved our distribution modeling capabilities, showing promise for dataset augmentation, image-to-image translation and feature learning. However, to model high-dimensional distributions, sequential training and stacked architectures are common, increasing the number of tunable hyper-parameters as well as the training time. Nonetheless, the sample complexity of the distance metrics remains one of the factors affecting GAN training. We first show that the recently proposed sliced Wasserstein distance has compelling sample complexity properties when compared to the Wasserstein distance. To further improve the sliced Wasserstein distance we then analyze its 'projection complexity' and develop the max-sliced Wasserstein distance which enjoys compelling sample complexity while reducing projection complexity, albeit necessitating a max estimation. We finally illustrate that the proposed distance trains GANs on high-dimensional images up to a resolution of 256x256 easily.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Max-Sliced Wasserstein Distance and Its Use for GANs

Deshpande

Sun

et al. 2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

104

View full text Add to dashboard Cite

show abstract

“…Self-training has been exploited in various tasks such as semi-supervised learning [25,21], domain adaptation [38,58], and noisy label learning [40,35]. [41,44,34,47,24,13] adopted adversarial training at feature level to learn domain-invariant features to reduce the discrepancy across domains. [18,8,27] applied adversarial training at the image level to make features invariant to illumination, color and other style factors.…”

Section: Related Workmentioning

confidence: 99%

Content-Consistent Matching for Domain Adaptive Semantic Segmentation

Kang

Liu

et al. 2020

Lecture Notes in Computer Science

122

View full text Add to dashboard Cite

This paper considers the adaptation of semantic segmentation from the synthetic source domain to the real target domain. Different from most previous explorations that often aim at developing adversarialbased domain alignment solutions, we tackle this challenging task from a new perspective, i.e., content-consistent matching (CCM). The target of CCM is to acquire those synthetic images that share similar distribution with the real ones in the target domain, so that the domain gap can be naturally alleviated by employing the content-consistent synthetic images for training. To be specific, we facilitate the CCM from two aspects, i.e., semantic layout matching and pixel-wise similarity matching. First, we use all the synthetic images from the source domain to train an initial segmentation model, which is then employed to produce coarse pixel-level labels for the unlabeled images in the target domain. With the coarse/accurate label maps for real/synthetic images, we construct their semantic layout matrixes from both horizontal and vertical directions and perform the matrixes matching to find out the synthetic images with similar semantic layout to real images. Second, we choose those predicted labels with high confidence to generate feature embeddings for all classes in the target domain, and further perform the pixel-wise matching on the mined layout-consistent synthetic images to harvest the appearance-consistent pixels. With the proposed CCM, only those content-consistent synthetic images are taken into account for learning the segmentation model, which can effectively alleviate the domain bias caused by those content-irrelevant synthetic images. Extensive experiments are conducted on two popular domain adaptation tasks, i.e., GTA5− →Cityscapes and SYNTHIA− →Cityscapes. Our CCM yields consistent improvements over the baselines and performs favorably against previous state-of-the-arts.

show abstract

“…[47] applies adversarial loss directly on learned segmentation features maps. Another examples are [34], [50], [4] and [54], [5].…”

Section: Related Workmentioning

confidence: 99%

KLIEP-based Density Ratio Estimation for Semantically Consistent Synthetic to Real Images Adaptation in Urban Traffic Scenes

Savkin

Tombari

2020

2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

Synthetic data has been applied in many deep learning based computer vision tasks. Limited performance of algorithms trained solely on synthetic data has been approached with domain adaptation techniques such as the ones based on generative adversarial framework. We demonstrate how adversarial training alone can introduce semantic inconsistencies in translated images. To tackle this issue we propose density prematching strategy using KLIEP-based density ratio estimation procedure. Finally, we show that aforementioned strategy improves quality of translated images of underlying method and their usability for the semantic segmentation task in the context of autonomous driving.

show abstract

Sliced Wasserstein Generative Models

Cited by 87 publications

References 16 publications

Max-Sliced Wasserstein Distance and Its Use for GANs

Max-Sliced Wasserstein Distance and Its Use for GANs

Content-Consistent Matching for Domain Adaptive Semantic Segmentation

KLIEP-based Density Ratio Estimation for Semantically Consistent Synthetic to Real Images Adaptation in Urban Traffic Scenes

Contact Info

Product

Resources

About