Method for learning cross-domain relations based on generative adversarial networks

Abstract

A generative adversarial networks-based or GAN-based method for learning cross-domain relations is disclosed. A provided architecture includes two coupled GANs: a first GAN learning a translation of images from domain A to domain B, and a second GAN learning a translation of images from domain B to domain A. A loop formed by the first GAN and the second GAN causes sample images to be reconstructed into an original domain after being translated into a target domain. Therefore, loss functions representing reconstruction losses of the images may be used to train generative models.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

The present application is based on, and claims priority from, Korean Patent Application Number 10-2017-0054474, filed Apr. 27, 2017, the disclosure of which is incorporated by reference herein in its entirety.

TECHNICAL FIELD

The present disclosure in some embodiments relates to a generative adversarial networks.

BACKGROUND

The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

Manners of interconnections from concepts to concepts, from objects to objects, and from persons to persons, etc. are ubiquitous cross-domain relations. Cross-domain relations are, in many cases, natural things happening to people. For example, we are aware of the relationship between English texts and their French translation and vice versa. In addition, we select the jacket of suit to wear in style coherent to pants and shoes.

The computer′ problem of whether it can obtain a humanlike capacity to associate domains of two different kinds of images may be reconstructed with a conditional image generation problem. In other words, finding a mapping function from one domain to another is analogous to being responsive to given images in one domain for generating corresponding images in the other domain. Conventionally, such image-to-image translation has been studied mainly by methods in which a training set composed of mutually paired images is used for learning the mapping between the input images and the output images. However, it is very troublesome to prepare paired training data, and the data are unusable in many tasks. The recent introduction of Generative Adversarial Networks (GAN) has led active attempts to apply the algorithm to conditional image generation.

FIG. 1 is a schematic diagram of GAN. GAN was introduced by Non-Patent Document 1, as a relatively new machine learning architecture for a neural network. A machine learning algorithm, GAN belongs to a part of unsupervised learning, and it is a new type of generation model used to generate images. The GAN concept introduces a discriminator network (D), to solve the problem of training a generator network (G). The word “adversarial” refers to two adversarial networks, i.e., “discriminator” and “generator.” As illustrated in FIG. 1, the generator tries to generate a more realistic image that can fool the discriminator, while the discriminator continuously adjust parameters to discriminate the images generated by the generator from the real image. In terms of game theory, these networks compete with each other in a zero-sum game.

CITATION LIST

• Non-patent document 1: Goodfellow, lan, et al. “Generative adversarial nets.” Advances in Neural Information Processing Systems. 2014.
• Non-patent document 2: Maas, A. L., Hannun, A. Y., and Ng, A. Y. Rectifier nonlinearities improve neural network acoustic models. In Proceedings of the 30th International Conference on Machine Learning (ICML), 2013.
• Non-patent document 3: Xu, B., Wang, N., T., Chen, and Li, M. Empirical evaluation of rectified activations in convolutional network. In arXiv preprint arXiv:1505:00853, 2015.

SUMMARY

The present disclosure in some embodiments provides a new GAN-based architecture for learning cross-domain relations between two sets of unlabeled images.

In accordance with some embodiments, the present disclosure provides a generative adversarial networks-based or GAN-based architecture in which a first GAN (generative adversarial networks) learns a translation of images in domain A to images in domain B, and a second GAN learns a translation of the images in domain B to the images in domain A. A closed loop formed by the first GAN and the second GAN causes sample images to be reconstructed into an original domain after being translated into an target domain. Therefore, loss functions representing reconstruction losses of the images may be used to learning generative models.

According to one aspect of the present disclosure, a generative adversarial networks-based or GAN-based system is provided for discovering cross-domain relations from two sets of unpaired sample images. The GAN-based system includes a first GAN configured to learn a first generative model for translating images in domain A to images in domain B, and a second GAN configured to learn a second generative model for translating the images in domain B to the images in domain A. The first GAN and the second GAN form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain. The first GAN and the second GAN use reconstruction losses of the sample images for learning the first generative model and the second generative model, the reconstruction losses indicating how well the translated sample image has been reconstructed into the original domain.

According to another aspect, a GAN-based system for discovering cross-domain relations from two sets of unpaired sample images includes a first generator configured to learn a first generative model for performing an image-to-image translation from a first domain to a second domain, a first discriminator configured to learn a first discriminative model for discriminating between images translated by the first generator and sample images representing the second domain, a second generator configured to learn a second generative model for performing an image-to-image translation from the second domain to the first domain, and a second discriminator configured to learn a second discriminative model for discriminating between images translated by the second generator and sample images representing the first domain. The first generator and the second generator form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain.

The first generator and the second generator respectively iteratively update the first generative model and the second generative model, respectively, so as to minimize reconstruction losses of the sample images and standard GAN generator losses of the sample images. Here, the reconstruction losses indicate how well the translated sample image has been reconstructed into the original domain, and the standard GAN generator losses indicate how realistic the translated images are in a target domain.

In yet another aspect, the present disclosure provides a method of discovering cross-domain relations from two sets of unpaired sample images. The method includes providing a first GAN to perform an unsupervised learning of a first generative model for translating images in domain A to images in domain B, and a second GAN to perform a unsupervised learning of a second generative model for translating the images in domain B to the images in domain A. Here, the first GAN and the second GAN form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain. The method uses reconstruction losses in to train the first generative model and the second generative model, the reconstruction losses indicating how well the translated sample image has been reconstructed into the original domain.

In yet another aspect, the present disclosure provides a method of discovering cross-domain relations from two sets of unpaired sample images. The method utilizes a first GAN to perform an unsupervised training of a first generative model for translating images in domain A to images in domain B, and a second GAN to perform a second unsupervised training of a second generative model for translating the images in domain B to the images in domain A. The first GAN and the second GAN is coupled to form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain. The method also includes using reconstruction losses in to train the first generative model and the second generative model, the reconstruction losses indicating how well the translated sample image has been reconstructed into the original domain.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic diagram of generative adversarial networks or GAN.

FIG. 2A is an exemplary diagram of a structure of a generator for receiving an image of size 64×64×3 and outputting a translated image of the same size, and FIG. 2B is an exemplary diagram of a discriminator for discriminating images of size 64×64×3.

FIG. 3 is a schematic diagram of a standard GAN model-based architecture for a relation discovery process.

FIG. 4 is a schematic diagram of a GAN architecture with a reconstruction loss.

FIG. 5A is a schematic diagram of cross-domain relation discovery models in simplified one-dimensional domains, illustrating the ideal mapping, FIG. 5B illustrates a failed standard GAN model mapping, and FIG. 5C illustrates a failed mapping by GAN with a reconstruction loss.

FIG. 6 is a schematic diagram of an overall GANs-based architecture for relations discovery, according to at least one embodiment of the present disclosure.

FIGS. 7A and 7B are exemplary schematic diagrams of methods of training generators and discriminators according to at least one embodiment.

FIGS. 8A to 8C are diagrams of the results of applying a GANs-based model according to at least one embodiment to an image-to-image translation.

DETAILED DESCRIPTION

Some embodiments of the present disclosure are described below with reference to the accompanying exemplary drawings. In the following description, like reference numerals would rather designate like elements, although the elements are shown in different drawings. Further, in the following description of some embodiments, a detailed description of known functions and configurations incorporated herein will be omitted for the purpose of clarity and for brevity and not to obscure the subject matter of the present disclosure.

The present disclosure in some embodiments provides a new GAN-based model that can learn cross-domain relations from unlabeled images. The disclosed model in some embodiments is trained with two sets of images without requiring images with explicit labels (i.e., paired examples) and obviates the need for a prior training. The disclosed model is capable of translating images from one domain to another domain by using learned cross-domain relations.

In short, the present disclosure couples two different GANs together. The respective GANs have generators for mapping the respective domains to their counterpart domains. The key intuition of the present disclosure is to constrain all images in one domain to be representable by images in the other domain. For example, when learning to generate a shoe image based on each handbag image, the disclosed model forces this generated image to be an image-based representation of the handbag image (and hence reconstructs the handbag image) through a reconstruction loss, and to be as close to the images in the shoe domain as possible through a GAN loss. These two properties used encourage bilateral one-to-one (rather than many-to-one or one-to-many) mapping between two domains.

The following describes a formularized definition of cross-domain relations and presents the problem of learning to discover such relations in two different domains. Standard GAN model (proposed by Non-patent document 1) and a similar variant model will be cited as to their applicability for this task, which has limitations to be addressed by the present GANs-based architecture for discovering cross-domain relations.

The cross-domain relation is mathematically defined as a function G_AB that maps elements from its domain A to elements in its codomain B, and as a function G_BA that is similarly defined. In fully unsupervised setting, functions G_AB and G_BA may be arbitrarily defined. To find a meaningful relation, a condition needs to be imposed on the relation of interest. Some embodiments constrain relation to be a one-to-one correspondence (i.e., bijective mapping), which means that G_AB is the inverse mapping of G_BA. In other words, the range of function G_AB, which is the complete set of all possible resulting values G_AB(x_A) for all x_A's in domain A, needs to be contained in domain B. Likewise, the range of function G_BA, which is the complete set of all possible resulting values G_BA(x_B) for all x_B's in domain B, needs to be contained in domain A.

The present disclosure relates these constraints to objective functions. Ideally, the equality G_BA∘G_AB(x_A)=x_A is satisfied, but this hard constraint is difficult to optimize and a relaxed soft constraint is more desirable in the view of optimization. For this reason, present disclosure provides a model which minimizes the distance d(G_BA∘G_AB(x_A), x_A), where any form of metric function (L₁, L₂, Huber loss, etc.) can be used. Similarly, the disclosed model is designed to minimize d(G_AB∘G_BA(x_B), x_B).

Guaranteeing that G_A maps to domain B is also very difficult to optimize. The present disclosure relaxes this constraint by minimizing two generative adversarial losses represented by Equation 1 and Equation 2.−_xA˜PA[ log D_B(G_AB(x_A))]  Equation 1−_xB˜PB[ log D_A(G_BA(x_B))]  Equation 2

Here, is an expectation function, and x_A and x_B are samples of domain A and B, respectively.

The following describes some GAN architectures to be trained with these loss functions. The description for GAN architectures below will use the following notation.

For example, a generator as denoted by G_A is represented by Equation 3 below. Subscripts denote the input and output domains and superscripts denote the input and output image sizes. A discriminator as denoted by D_B is represented by Equation 4 below. In Equation 4 that is [0, 1], subscript B denotes that it discriminates images in domain B. Equations 3 and 4 assume that domains A and B have images of size 64×64×3.G_AB:_A^64×64×3→_B^64×64×3  Equation 3D_B:_B^64×64×3→[0,1]  Equation 4

Notations G_BA and D_A are used for the other generator and the other discriminator similar to Equations 3 and 4.

FIG. 2A is an exemplary diagram of a structure of a generator for receiving an image of size 64×64×3 and outputting a translated image of the same size, and FIG. 2B is an exemplary diagram of a discriminator for discriminating images of size 64×64×3.

Each generator can be seen as an encoder-decoder pair. The generator may have an encoder section composed of convolution layers with 4×4 filters, each followed by a leaky ReLU (Rectified Linear Unit). The generator may also have a decoder section composed of deconvolution layers with 4×4 filters, followed by a ReLU, and it finally outputs a target domain image of size 64×64×3. The numbers of the convolution and deconvolution layers may range from four to five, depending on the domain. In addition, the dimensions of the convolution and deconvolution layers may be appropriately modified. For detailed descriptions of ReLU and leaky ReLU, Non-Patent Documents 2 and 3 may be consulted.

The discriminator is similar to the encoder section of the generator. Besides the convolution layers and leaky ReLUs, the discriminator has an additional convolution layer with 4×4 filters, and a final sigmoid for outputting a scalar output between [0, 1].

FIG. 3 is a schematic diagram of a standard GAN model-based architecture for a relation discovery task.

A standard GAN is supposed to take a random Gaussian noise z, encode it into hidden features h and generate images such as MNIST digits. A slight modification needs to be made to this model to fit the relation discovery task: the modified model takes in an image as input instead of noise, as shown in FIG. 3.

FIG. 4 is a schematic diagram of a GAN architecture with a reconstruction loss.

Different from the architecture of FIG. 3, which only learns one mapping from domain A to domain B, the architecture of FIG. 4 is provided with a second generator for mapping from domain B back into domain A. Another addition made to the architecture of FIG. 4 is a reconstruction loss term that compares the input image with the reconstructed image. In such GAN architecture with these additional changes, each generator (G_AB, G_BA) can learn the mapping from its input domain to output domain and discover relations between the two domains.

Generator G_AB translates input image x_A from domain A into x_AB in domain B as represented by Equation 5. The generated image x_AB is then translated into a domain-A image x_ABA to match the original input image x_A as represented by Equation 6. Various forms of distance functions, such as MSE, cosine distance, and hinge-loss, can be used as the reconstruction loss d as represented by Equation 7. The translated output x_AB is then scored by discriminator D_B which compares the translated output x_AB to a real sample x_B in domain B.x_AB=G_AB(x_A)  Equation 5x_ABA=G_BA(x_AB)=G_BA∘G_AB(x_A)  Equation 6L_CONSTA=d(G_BA∘G_AB(x_A),x_A)  Equation 7L_GANB=−_xA˜PA[ log D_B(G_AB(x_A))]  Equation 8

Generator G_AB receives two types of losses as in Equation 9. A first type is a standard GAN generator loss L_GANB as represented by Equation 8 that measures how realistic the generated image is in domain B, and a second type is a reconstruction loss L_CONSTA as represented by Equation 7 that measures how well the original input is reconstructed after a sequence of two generations. Discriminator D_B receives the standard GAN discriminator loss of Equation 10.T_GAB=L_GANB+L_CONSTA  Equation 9L_DB=−_xB˜PB[ log D_B(x_B)]−_xA˜PA[ log(1−D_B(G_AB(x)))]  Equation 10

During training, the generator G_AB learns the mapping from domain A to domain B under two relaxed constraints: that domain A maps to domain B, and that the mapping on domain B is reconstructed into domain A. However, this model lacks a constraint on mapping from B to A, and these two conditions alone does not guarantee a cross-domain relation because the mapping satisfying these constraints is one-directional. In other words, the mapping is an injection, not bijection, and one-to-one correspondence is not guaranteed.

Considered are the two possibly multi-modal image domains A and B. FIG. 5 illustrates the two multi-modal data domains on a simplified one-dimensional representation. FIG. 5A shows the ideal mapping from input domain A to domain B, where each mode of data is mapped to a separate mode in the target domain. In contrast, FIG. 5B shows the mode collapse problem, a prevalent phenomenon in GANs, where data from multiple modes of a domain map to a single mode of a different domain. For instance, this case is where the mapping G_AB maps images of cars in two different orientations into the same mode of face images.

In some sense, the architecture of FIG. 4 with the addition of a reconstruction loss to a standard GAN is an attempt to remedy the mode collapse problem. FIG. 5C shows that two domain-A modes are matched with the same domain-B mode, but the same domain-B mode can only direct to one of the two domain-A modes. Although the additional reconstruction loss L_CONSTA forces the reconstructed sample to match the original, this change only leads to a similar symmetric problem (FIG. 5C). The reconstruction loss leads to an oscillation between the two symmetrical states and does not resolve mode-collapsing.

With reference to FIG. 6, a GANs-based model called DiscoGAN for relation discovery in accordance with at least one embodiment of the present disclosure will be described. FIG. 6 is a conceptual diagram of an overall GAN-based architecture for relation discovery, according to at least one embodiment of the present disclosure.

The architecture of FIG. 6 includes two coupled GAN models, in which two GAN models are coupled to offer two reconstruction losses. The two coupled models, each learns the mapping from one domain to another, and also the reverse mapping back to the one domain for reconstruction. The two GAN models are trained together simultaneously. As FIG. 6 illustrates, a generator G_AB is engaged in two translation tasks and generator G_BA is engaged in two translation tasks. In other words, the generator G_AB translates the sample images x_A in domain A to images x_AB in domain B and also translates (i.e., reconstructs) the translated images x_BA, which are generated by the generator G_BA, to images x_BAB in domain B. Therefore, the generator G_AB applies the parameters used for translating the sample images x_A in domain A to the images x_BA in domain B, as the common parameters for reconstructing the translated images x_BA into domain B.

The generated images x_AB and x_BA from generators G_AB and G_BA are each fed into separate discriminators D_B and D_A, respectively.

The model of FIG. 6 has at least one key difference from the previous model of FIG. 4 that the sample images (x_A, x_B) from both domains are translated and then are reconstructed into the original domains and that there are two reconstruction losses: L_CONSTA and L_CONSTB.

As a result of coupling two GAN models, the total generator loss is the sum of GAN losses and reconstruction losses for the respective partial models as presented by Equation 11. Similarly, the total discriminator loss L_D is a sum of discriminator losses of the two discriminators D_A and D_B which discriminate real from fake images in domain A and domain B as presented by Equation 12.

$\begin{array}{cc}\begin{array}{c}{L}_G=L_{G_{\mathrm{AB}}}+L_{G_{\mathrm{BA}}}\\ =L_{{\mathrm{GAN}}_B}+L_{{\mathrm{CONST}}_A}+L_{{\mathrm{GAN}}_A}+L_{{\mathrm{CONST}}_B}\end{array}& \mathrm{Equation}11\end{array}$ L _D =L _{D A} +L _{D B}   Equation 12

Now, the present model is constrained by two L_GAN losses and two L_CONST losses. This achieves a bijective mapping, enabling the discovery of a one-to-one correspondence defined by the present disclosure as cross-domain relations.

The generators and the discriminators are alternately and repeatedly trained. For example, in one step, discriminators D_B and D_A are updated (e.g., changing the weight of the value passed between the deconvolution layers) toward minimizing the discriminator losses, and in the subsequent step, generators G_AB and G_BA are updated (e.g., changing the weight of the value passed between the deconvolution layers) toward minimizing the total generator loss determined by the updated discriminators and expressed by Equation 11.

FIGS. 7A and 7B are exemplary schematic diagrams of methods of training generators and discriminators according to at least one embodiment. Illustrative methods of training the GAN-based system shown in FIG. 6 to discover cross-domain relations will be described by using 1000 sample images representing domain A and 1000 images representing domain B.

First, referring to FIG. 7A, a method of training the generators will be described. Upon receiving sample images x_A in domain A, generator G_AB outputs translated images x_AB to domain B. The translated images x_AB are input to generator G_BA where they are reconstructed into the original one (i.e., domain A) of domains A and B. Derived from the translated images denoted by x_ABA and the original sample images x_A are reconstruction losses L_CONSTA. The above process is repeated over the representative sample images of domain A as much as, for example, 100 sample images. Generator G_AB updates its parameters toward reducing the sum or average of reconstruction losses L_CONSTA of the 100 sample images x_A. In the same way, reconstruction losses L_CONSTB of the 100 sample images x_B representing domain B are obtained, and the parameters of generator G_BA are updated toward reducing the sum or average of reconstruction losses L_CONSTB.

Referring to FIG. 7B, a method of training the discriminators will be described. Upon receiving 100 sample images x_A representative of domain A and 100 translated (by generator G_BA updated in the previous stage) domain-A images x_BA, discriminator D_A generates 200 probabilities. Parameters of discriminator D_A are updated toward reducing the sum or average of 100 probabilities. Likewise, discriminator D_B, upon receiving 100 sample images x_B representative of domain B and 100 translated (by generator G_AB updated in the previous stage) domain-B images x_AB, generates 200 probabilities. Parameters of discriminator D_A are updated toward reducing the sum or average of 100 probabilities. Parameters of discriminator D_B are updated toward reducing the sum or average of 100 probabilities.

Calculations for the provision of the sample images, the reconstruction losses and other terms may be carried out by a training module (not shown) having one or more processors and storages. The generators and discriminators are alternately and repeatedly trained, when the set of sample images used in the earlier training stage can be reused in the next training stage, although it is efficient for training to use a new set of sample images.

FIGS. 8A to 8C are diagrams of the results of applying a GANs-based model according to at least one embodiment to an image-to-image translation. In some embodiments, the present architecture is capable of simultaneous training of two reliable image-to-image translators respectively performing cross-domain translations of images by using inputs of two sets of unlabeled images (sets of sample images not mutually paired) representing image domains, respectively, as illustrated in FIG. 8A. For example, of two generators, one G_AB may be responsive to handbag images for operating as another image-to-image translator that generates shoe images coherent in style to the handbag images, respectively, as illustrated in FIG. 8B. The other generator G_BA may be responsive to shoe images for operating as an image-to-image translator that generates handbag images coherent in style to the shoe images, respectively, as illustrated in FIG. 8C.

Neural networks according to the architecture provided by some embodiments of the present disclosure can be trained to discover relations between two different domains from two sets of unlabeled images. Using the discovered relations, the neural networks can transmit styles from one domain to the other domain successfully, while maintaining the key attributes such as orientations, facial identities, etc.

The provided architecture resolves all the issues of mode collapse and mode oscillation.

Architecture provided by the present disclosure is capable of simultaneous training of two reliable image-to-image translators respectively performing cross-domain translations of images by using inputs of two sets of unlabeled images representing image domains, respectively, and it can perform various image translation processing.

The aforementioned GAN-based method of discovering cross-domain relations may be implemented as computer-readable codes in a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording devices storing data that can be read by a computer system. In other words, the computer-readable recording medium includes a magnetic recording medium (e.g., ROM, floppy disks, hard disks, etc.), and a storage medium such as an optically readable media (e.g., CD-ROM, DVD, etc.). Further, the computer-readable recording medium can be distributed in computer systems connected via a network, and computer-readable codes can be stored and executed in a distributed mode.

Although exemplary embodiments of the present disclosure have been described for illustrative purposes, those skilled in the art will appreciate that various modifications, additions and substitutions are possible, without departing from the various characteristics of the disclosure. Therefore, exemplary embodiments of the present disclosure have been described for the sake of brevity and clarity. Accordingly, one of ordinary skill would understand the scope of the disclosure is not limited by the explicitly described above embodiments but by the claims and equivalents thereof. The scope of protection of the present embodiments is to be construed by the following claims, and all the technical ideas within the scope of their equivalents should be construed as being included in the scope of the embodiments.

Claims

1. A generative adversarial networks-based or GAN-based system for discovering cross-domain relations from two sets of unpaired sample images, the GAN-based system comprising: a first GAN configured to learn a first generative model for translating images from domain A to domain B; anda second GAN configured to learn a second generative model for translating images from domain B to domain A,wherein the first GAN and the second GAN form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain, andwherein the first GAN and the second GAN use reconstruction losses of the sample images for learning the first generative model and the second generative model, the reconstruction losses indicating how well the translated sample image has been reconstructed into the original domain.

2. The GAN-based system of claim 1, wherein the reconstruction loss is calculated by a distance function representing a difference between the sample image and the reconstructed image.

3. The GAN-based system of claim 1, wherein the first GAN and the second GAN each comprises: a generator configured to learn a generative model for translating images in the original domain in the target domain, anda discriminator configured to learn a discriminative model for discriminating between translated images generated by the generator and sample images in the target domain.

4. The GAN-based system of claim 3, wherein the first GAN and the second GAN iteratively update the first generative model and the second generative model, respectively, so as to minimize the reconstruction losses and standard GAN generator losses indicating how realistic the translated images are in the target domain.

5. The GAN-based system of claim 4, wherein the standard GAN generator loss is defined for each GAN by −xA˜PA[ log DB(GAB(xA))] and −xB˜PB[ log DA(GBA(xB))]

6. The GAN-based system of claim 3, wherein the generator comprises: an encoder including a plurality of convolution layers; anda decoder including a plurality of deconvolution layers.

7. The GAN-based system of claim 3, wherein the discriminator comprises: an encoder including a plurality of deconvolution layers and a sigmoid.

8. A generative adversarial networks-based or GAN-based system for discovering cross-domain relations from two sets of unpaired sample images, the GAN-based system comprising: a first generator configured to learn a first generative model for performing an image-to-image translation from a first domain to a second domain;a first discriminator configured to learn a first discriminative model for discriminating between images translated by the first generator and sample images representing the second domain;a second generator configured to learn a second generative model for performing an image-to-image translation from the second domain to the first domain; anda second discriminator configured to learn a second discriminative model for discriminating between images translated by the second generator and sample images representing the first domain,wherein the first generator and the second generator form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain, andwherein the first generator and the second generator iteratively update the first generative model and the second generative model, respectively, so as to minimize reconstruction losses of the sample images and standard GAN generator losses of the sample images, the reconstruction losses indicating how well the translated sample image has been reconstructed into the original domain, and the standard GAN generator losses indicating how realistic the translated images are in a target domain.

9. The GAN-based system of claim 8, wherein the reconstruction loss is calculated by a distance function representing a difference between an original sample image and a corresponding reconstructed image.

10. A method of discovering cross-domain relations from two sets of unpaired sample images, the method comprising: providing a first GAN (generative adversarial networks) to perform an unsupervised learning of a first generative model for translating images from domain A to domain B, and a second GAN to perform a second unsupervised learning of a second generative model for translating the images from domain B to domain A;configuring the first GAN and the second GAN to form a loop such that the sample images in an original domain are translated into a target domain and then translated sample images are reconstructed into the original domain; andusing reconstruction losses to train the first generative model and the second generative model, the reconstruction losses indicating how well the translated sample image has been reconstructed into the original domain.