A Sliced Wasserstein Loss for Neural Texture Synthesis

Heitz, Eric; Vanhoey, Kenneth; Chambon, Thomas; Belcour, Laurent

doi:10.1109/cvpr46437.2021.00929

Cited by 31 publications

(25 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The sliced Wasserstein loss has been shown to be a good statistical metric [HVCB21] to compare deep feature maps, but does not allow comparing feature maps with different numbers of samples (pixels) trivially. The "tag" trick introduced in the original paper cannot be applied to our goal for multi-target transfer.…”

Section: Spatial Control and Multi-target Transfermentioning

confidence: 99%

“…For example, Gatys et al [GEB15,GEB16] leverage a pre-trained VGG neural network [SZ15] to guide style transfer, using the Gram Matrix of extracted deep features from the image as their statistical representation. Heitz et al [HVCB21] described an alternative sliced Wasserstein loss as a more complete statistical description of extracted features. Different approaches proposed to train a neural network to transfer style of images or synthesize textures in a single forward operation [JAFF16, ULVL16, HB17, ZZB * 18].…”

Section: Style Transfermentioning

confidence: 99%

See 1 more Smart Citation

Controlling Material Appearance by Examples

Hašan

Guerrero

et al. 2022

Computer Graphics Forum

View full text Add to dashboard Cite

Figure 1: We propose a method to control and improve the appearance of an existing material (left) by transferring the appearance of materials in one or multiple target photo(s) (center) to the existing material. The augmented material (right) combines the coarse structure from the original material with the fine-scale appearance of the target(s) and preserve the input tileability. Our method can also transfer appearance from materials from different types by spatial control. This enables a simple workflow to make existing materials more realistic using readily-available images or photos.

show abstract

Section: Spatial Control and Multi-target Transfermentioning

confidence: 99%

Section: Style Transfermentioning

confidence: 99%

Controlling Material Appearance by Examples

Hašan

Guerrero

et al. 2022

Computer Graphics Forum

View full text Add to dashboard Cite

show abstract

“…The sliced Wasserstein (SW) distance has exhibited outstanding merit for training deep generative networks [9,46]. Recently, SW loss has been successfully applied in texture synthesis [17], image enhancement [8] and etc. Here, we also use SW loss L SW to optimize SelfDZSR, the detailed description will be given in the suppl.…”

Section: Learning Objectivementioning

confidence: 99%

Self-Supervised Learning for Real-World Super-Resolution from Dual Zoomed Observations

Zhang¹,

Ruohao²,

Zhang³

et al. 2022

Preprint

View full text Add to dashboard Cite

In this paper, we consider two challenging issues in reference-based super-resolution (RefSR), (i) how to choose a proper reference image, and (ii) how to learn real-world RefSR in a self-supervised manner. Particularly, we present a novel self-supervised learning approach for real-world image SR from observations at dual camera zooms (Self-DZSR). For the first issue, the more zoomed (telephoto) image can be naturally leveraged as the reference to guide the SR of the lesser zoomed (short-focus) image. For the second issue, SelfDZSR learns a deep network to obtain the SR result of short-focal image and with the same resolution as the telephoto image. For this purpose, we take the telephoto image instead of an additional high-resolution image as the supervision information and select a patch from it as the reference to super-resolve the corresponding short-focus image patch. To mitigate the effect of various misalignment between the short-focus low-resolution (LR) image and telephoto ground-truth (GT) image, we design a degradation model and map the GT to a pseudo-LR image aligned with GT. Then the pseudo-LR and LR image can be fed into the proposed adaptive spatial transformer networks (AdaSTN) to deform the LR features. During testing, SelfDZSR can be directly deployed to super-solve the whole short-focus image with the reference of telephoto image. Experiments show that our method achieves better quantitative and qualitative performance against state-of-the-arts. The code and pre-trained models will be publicly available.

show abstract

“…They avoid computing OT problems in high dimensional space by projecting the distributions onto random 1-dimensional lines, and solving a 1-dimensional OT problem each time. Due to their simplicity and efficiency, this class of approaches have been applied to several fields in computer vision, such as generative modelling (Deshpande et al, 2018) and texture synthesis (Heitz et al, 2021).…”

Section: B Existing Computational Approaches Of Ot Problemmentioning

confidence: 99%

Partial Wasserstein Adversarial Network for Non-rigid Point Set Registration

Wang¹,

Xue²,

Ling³

et al. 2022

Preprint

View full text Add to dashboard Cite

Given two point sets, the problem of registration is to recover a transformation that matches one set to the other. This task is challenging due to the presence of the large number of outliers, the unknown non-rigid deformations and the large sizes of point sets. To obtain strong robustness against outliers, we formulate the registration problem as a partial distribution matching (PDM) problem, where the goal is to partially match the distributions represented by point sets in a metric space. To handle large point sets, we propose a scalable PDM algorithm by utilizing the efficient partial Wasserstein-1 (PW) discrepancy. Specifically, we derive the Kantorovich-Rubinstein duality for the PW discrepancy, and show its gradient can be explicitly computed. Based on these results, we propose a partial Wasserstein adversarial network (PWAN), which is able to approximate the PW discrepancy by a neural network, and minimize it by gradient descent. In addition, it also incorporates an efficient coherence regularizer for non-rigid transformations to avoid unrealistic deformations. We evaluate PWAN on practical point set registration tasks, and show that the proposed PWAN is robust, scalable and performs more favorably than the state-of-the-art methods.

show abstract

A Sliced Wasserstein Loss for Neural Texture Synthesis

Cited by 31 publications

References 15 publications

Controlling Material Appearance by Examples

Controlling Material Appearance by Examples

Self-Supervised Learning for Real-World Super-Resolution from Dual Zoomed Observations

Partial Wasserstein Adversarial Network for Non-rigid Point Set Registration

Contact Info

Product

Resources

About