Abstract: One example method involves operations for receiving a request to transform an input image into a target image. Operations further include providing the input image to a machine learning model trained to adapt images. Training the machine learning model includes accessing training data having a source domain of images and a target domain of images with a target style. Training further includes using a pre-trained generative model to generate an adapted source domain of adapted images having the target style. The adapted source domain is generated by determining a rate of change for parameters of the target style, generating weighted parameters by applying a weight to each of the parameters based on their respective rate of change, and applying the weighted parameters to the source domain. Additionally, operations include using the machine learning model to generate the target image by modifying parameters of the input image using the target style.

Abstract: An improved system architecture uses a pipeline including a Generative Adversarial Network (GAN) including a generator neural network and a discriminator neural network to generate an image. An input image in a first domain and information about a target domain are obtained. The domains correspond to image styles. An initial latent space representation of the input image is produced by encoding the input image. An initial output image is generated by processing the initial latent space representation with the generator neural network. Using the discriminator neural network, a score is computed indicating whether the initial output image is in the target domain. A loss is computed based on the computed score. The loss is minimized to compute an updated latent space representation. The updated latent space representation is processed with the generator neural network to generate an output image in the target domain.

Abstract: Systems and methods generate a filtering function for editing an image with reduced attribute correlation. An image editing system groups training data into bins according to a distribution of a target attribute. For each bin, the system samples a subset of the training data based on a pre-determined target distribution of a set of additional attributes in the training data. The system identifies a direction in the sampled training data corresponding to the distribution of the target attribute to generate a filtering vector for modifying the target attribute in an input image, obtains a latent space representation of an input image, applies the filtering vector to the latent space representation of the input image to generate a filtered latent space representation of the input image, and provides the filtered latent space representation as input to a neural network to generate an output image with a modification to the target attribute.

Abstract: Various disclosed embodiments are directed to inpainting one or more portions of a target image based on merging (or selecting) one or more portions of a warped image with (or from) one or more portions of an inpainting candidate (e.g., via a learning model). This, among other functionality described herein, resolves the inaccuracies of existing image inpainting technologies.

Abstract: This disclosure describes methods, non-transitory computer readable storage media, and systems that generate synthetized digital images using class-specific generators for objects of different classes. The disclosed system modifies a synthesized digital image by utilizing a plurality of class-specific generator neural networks to generate a plurality of synthesized objects according to object classes identified in the synthesized digital image. The disclosed system determines object classes in the synthesized digital image such as via a semantic label map corresponding to the synthesized digital image. The disclosed system selects class-specific generator neural networks corresponding to the classes of objects in the synthesized digital image. The disclosed system also generates a plurality of synthesized objects utilizing the class-specific generator neural networks based on contextual data associated with the identified objects.

Abstract: A panoptic labeling system includes a modified panoptic labeling neural network (“modified PLNN”) that is trained to generate labels for pixels in an input image. The panoptic labeling system generates modified training images by combining training images with mask instances from annotated images. The modified PLNN determines a set of labels representing categories of objects depicted in the modified training images. The modified PLNN also determines a subset of the labels representing categories of objects depicted in the input image. For each mask pixel in a modified training image, the modified PLNN calculates a probability indicating whether the mask pixel has the same label as an object pixel. The modified PLNN generates a mask label for each mask pixel, based on the probability. The panoptic labeling system provides the mask label to, for example, a digital graphics editing system that uses the labels to complete an infill operation.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media that implement an inpainting framework having computer-implemented machine learning models to generate high-resolution inpainting results. For instance, in one or more embodiments, the disclosed systems generate an inpainted digital image utilizing a deep inpainting neural network from a digital image having a replacement region. The disclosed systems further generate, utilizing a visual guide algorithm, at least one deep visual guide from the inpainted digital image. Using a patch match model and the at least one deep visual guide, the disclosed systems generate a plurality of modified digital images from the digital image by replacing the region of pixels of the digital image with replacement pixels. Additionally, the disclosed systems select, utilizing an inpainting curation model, a modified digital image from the plurality of modified digital images to provide to a client device.

Abstract: Certain aspects involve video inpainting in which content is propagated from a user-provided reference video frame to other video frames depicting a scene. One example method includes one or more processing devices that performs operations that include accessing a scene depicting a reference object that includes an annotation identifying a target region to be modified in one or more video frames. The operations also includes computing a target motion of a target pixel that is subject to a motion constraint. The motion constraint is based on a three-dimensional model of the reference object. Further, operations include determining color data of the target pixel to correspond to the target motion. The color data includes a color value and a gradient. Operations also include determining gradient constraints using gradient values of neighbor pixels. Additionally, the processing devices updates the color data of the target pixel subject to the gradient constraints.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media that generate inpainted digital images utilizing class-specific cascaded modulation inpainting neural network. For example, the disclosed systems utilize a class-specific cascaded modulation inpainting neural network that includes cascaded modulation decoder layers to generate replacement pixels portraying a particular target object class. To illustrate, in response to user selection of a replacement region and target object class, the disclosed systems utilize a class-specific cascaded modulation inpainting neural network corresponding to the target object class to generate an inpainted digital image that portrays an instance of the target object class within the replacement region.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media that generate inpainted digital images utilizing a cascaded modulation inpainting neural network. For example, the disclosed systems utilize a cascaded modulation inpainting neural network that includes cascaded modulation decoder layers. For example, in one or more decoder layers, the disclosed systems start with global code modulation that captures the global-range image structures followed by an additional modulation that refines the global predictions. Accordingly, in one or more implementations, the image inpainting system provides a mechanism to correct distorted local details. Furthermore, in one or more implementations, the image inpainting system leverages fast Fourier convolutions block within different resolution layers of the encoder architecture to expand the receptive field of the encoder and to allow the network encoder to better capture global structure.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for utilizing a critical edge detection neural network and a geometric model to determine camera parameters from a single digital image. In particular, in one or more embodiments, the disclosed systems can train and utilize a critical edge detection neural network to generate a vanishing edge map indicating vanishing lines from the digital image. The system can then utilize the vanishing edge map to more accurately and efficiently determine camera parameters by applying a geometric model to the vanishing edge map. Further, the system can generate ground truth vanishing line data from a set of training digital images for training the critical edge detection neural network.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for combining digital images. In particular, in one or more embodiments, the disclosed systems combine latent codes of a source digital image and a target digital image utilizing a blending network to determine a combined latent encoding and generate a combined digital image from the combined latent encoding utilizing a generative neural network. In some embodiments, the disclosed systems determine an intersection face mask between the source digital image and the combined digital image utilizing a face segmentation network and combine the source digital image and the combined digital image utilizing the intersection face mask to generate a blended digital image.

Abstract: An image inpainting system is described that receives an input image that includes a masked region. From the input image, the image inpainting system generates a synthesized image that depicts an object in the masked region by selecting a first code that represents a known factor characterizing a visual appearance of the object and a second code that represents an unknown factor characterizing the visual appearance of the object apart from the known factor in latent space. The input image, the first code, and the second code are provided as input to a generative adversarial network that is trained to generate the synthesized image using contrastive losses. Different synthesized images are generated from the same input image using different combinations of first and second codes, and the synthesized images are output for display.

Abstract: An item recommendation system receives a set of recommendable items and a request to select, from the set of recommendable items, a contrast group. The item recommendation system selects a contrast group from the set of recommendable items by applying a image modification model to the set of recommendable items. The image modification model includes an item selection model configured to determine an unbiased conversion rate for each item of the set of recommendable items and select a recommended item from the set of recommendable items having a greatest unbiased conversion rate. The image modification model includes a contrast group selection model configured to select, for the recommended item, a contrast group comprising the recommended item and one or more contrast items. The item recommendation system transmits the contrast group responsive to the request.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for latent-based editing of digital images using a generative neural network. In particular, in one or more embodiments, the disclosed systems perform latent-based editing of a digital image by mapping a feature tensor and a set of style vectors for the digital image into a joint feature style space. In one or more implementations, the disclosed systems apply a joint feature style perturbation and/or modification vectors within the joint feature style space to determine modified style vectors and a modified feature tensor. Moreover, in one or more embodiments the disclosed systems generate a modified digital image utilizing a generative neural network from the modified style vectors and the modified feature tensor.

Abstract: Methods, systems, and non-transitory computer readable media are disclosed for intelligently enhancing details in edited images. The disclosed system iteratively updates residual detail latent code for segments in edited images where detail has been lost through the editing process. More particularly, the disclosed system enhances an edited segment in an edited image based on details in a detailed segment of an image. Additionally, the disclosed system may utilize a detail neural network encoder to project the detailed segment and a corresponding segment of the edited image into a residual detail latent code. In some embodiments, the disclosed system generates a refined edited image based on the residual detail latent code and a latent vector of the edited image.

Abstract: Embodiments of the present invention provide systems, methods, and computer storage media for generating an animation of a talking head from an input audio signal of speech and a representation (such as a static image) of a head to animate. Generally, a neural network can learn to predict a set of 3D facial landmarks that can be used to drive the animation. In some embodiments, the neural network can learn to detect different speaking styles in the input speech and account for the different speaking styles when predicting the 3D facial landmarks. Generally, template 3D facial landmarks can be identified or extracted from the input image or other representation of the head, and the template 3D facial landmarks can be used with successive windows of audio from the input speech to predict 3D facial landmarks and generate a corresponding animation with plausible 3D effects.

Abstract: This disclosure describes methods, non-transitory computer readable storage media, and systems that generate synthetized digital images via multi-resolution generator neural networks. The disclosed system extracts multi-resolution features from a scene representation to condition a spatial feature tensor and a latent code to modulate an output of a generator neural network. For example, the disclosed systems utilizes a base encoder of the generator neural network to generate a feature set from a semantic label map of a scene. The disclosed system then utilizes a bottom-up encoder to extract multi-resolution features and generate a latent code from the feature set. Furthermore, the disclosed system determines a spatial feature tensor by utilizing a top-down encoder to up-sample and aggregate the multi-resolution features. The disclosed system then utilizes a decoder to generate a synthesized digital image based on the spatial feature tensor and the latent code.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods that implement a dual-branched neural network architecture to harmonize composite images. For example, in one or more implementations, the transformer-based harmonization system uses a convolutional branch and a transformer branch to generate a harmonized composite image based on an input composite image and a corresponding segmentation mask. More particularly, the convolutional branch comprises a series of convolutional neural network layers followed by a style normalization layer to extract localized information from the input composite image. Further, the transformer branch comprises a series of transformer neural network layers to extract global information based on different resolutions of the input composite image.

Abstract: Embodiments are disclosed for generative image congealing which provides an unsupervised learning technique that learns transformations of real data to improve the image quality of GANs trained using that image data. In particular, in one or more embodiments, the disclosed systems and methods comprise generating, by a spatial transformer network, an aligned real image for a real image from an unaligned real dataset, providing, by the spatial transformer network, the aligned real image to an adversarial discrimination network to determine if the aligned real image resembles aligned synthetic images generated by a generator network, and training, by a training manager, the spatial transformer network to learn updated transformations based on the determination of the adversarial discrimination network.