Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for accurately and flexibly generating modified digital images utilizing a novel swapping autoencoder that incorporates scene layout. In particular, the disclosed systems can receive a scene layout map that indicates or defines locations for displaying specific digital content within a digital image. In addition, the disclosed systems can utilize the scene layout map to guide combining portions of digital image latent code to generate a modified digital image with a particular textural appearance and a particular geometric structure defined by the scene layout map. Additionally, the disclosed systems can utilize a scene layout map that defines a portion of a digital image to modify by, for instance, adding new digital content to the digital image, and can generate a modified digital image depicting the new digital content.

Abstract: The present disclosure relates to an image composite system that employs a generative adversarial network to generate realistic composite images. For example, in one or more embodiments, the image composite system trains a geometric prediction neural network using an adversarial discrimination neural network to learn warp parameters that provide correct geometric alignment of foreground objects with respect to a background image. Once trained, the determined warp parameters provide realistic geometric corrections to foreground objects such that the warped foreground objects appear to blend into background images naturally when composited together.

Abstract: An improved system architecture uses a Generative Adversarial Network (GAN) including a specialized generator neural network to generate multiple resolution output images. The system produces a latent space representation of an input image. The system generates a first output image at a first resolution by providing the latent space representation of the input image as input to a generator neural network comprising an input layer, an output layer, and a plurality of intermediate layers and taking the first output image from an intermediate layer, of the plurality of intermediate layers of the generator neural network. The system generates a second output image at a second resolution different from the first resolution by providing the latent space representation of the input image as input to the generator neural network and taking the second output image from the output layer of the generator neural network.

Abstract: Systems and methods train an encoder neural network for fast and accurate projection into the latent space of a Generative Adversarial Network (GAN). The encoder is trained by providing an input training image to the encoder and producing, by the encoder, a latent space representation of the input training image. The latent space representation is provided as input to the GAN to generate a generated training image. A latent code is sampled from a latent space associated with the GAN and the sampled latent code is provided as input to the GAN. The GAN generates a synthetic training image based on the sampled latent code. The sampled latent code is provided as input to the encoder to produce a synthetic training code. The encoder is updated by minimizing a loss between the generated training image and the input training image, and the synthetic training code and the sampled latent code.

Abstract: Systems and methods combine an input image with an edited image generated using a generator neural network to preserve detail from the original image. A computing system provides an input image to a machine learning model to generate a latent space representation of the input image. The system provides the latent space representation to a generator neural network to generate a generated image. The system generates multiple scale representations of the input image, as well as multiple scale representations of the generated image. The system generates a first combined image based on first scale representations of the images and a first value. The system generates a second combined image based on second scale representations of the images and a second value. The system blends the first combined image with the second combined image to generate an output image.

Abstract: Systems and methods train and apply a specialized encoder neural network for fast and accurate projection into the latent space of a Generative Adversarial Network (GAN). The specialized encoder neural network includes an input layer, a feature extraction layer, and a bottleneck layer positioned after the feature extraction layer. The projection process includes providing an input image to the encoder and producing, by the encoder, a latent space representation of the input image. Producing the latent space representation includes extracting a feature vector from the feature extraction layer, providing the feature vector to the bottleneck layer as input, and producing the latent space representation as output. The latent space representation produced by the encoder is provided as input to the GAN, which generates an output image based upon the latent space representation. The encoder is trained using specialized loss functions including a segmentation loss and a mean latent loss.

Abstract: Systems and methods dynamically adjust an available range for editing an attribute in an image. An image editing system computes a metric for an attribute in an input image as a function of a latent space representation of the input image and a filtering vector for editing the input image. The image editing system compares the metric to a threshold. If the metric exceeds the threshold, then the image editing system selects a first range for editing the attribute in the input image. If the metric does not exceed the threshold, a second range is selected. The image editing system causes display of a user interface for editing the input image comprising an interface element for editing the attribute within the selected range.

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: An improved system architecture uses a pipeline including an encoder and a Generative Adversarial Network (GAN) including a generator neural network to generate edited images with improved speed, realism, and identity preservation. The encoder produces an initial latent space representation of an input image by encoding the input image. The generator neural network generates an initial output image by processing the initial latent space representation of the input image. The system generates an optimized latent space representation of the input image using a loss minimization technique that minimizes a loss between the input image and the initial output image. The loss is based on target perceptual features extracted from the input image and initial perceptual features extracted from the initial output image. The system outputs the optimized latent space representation of the input image for downstream use.

Abstract: Systems and methods seamlessly blend edited and unedited regions of an image. A computing system crops an input image around a region to be edited. The system applies an affine transformation to rotate the cropped input image. The system provides the rotated cropped input image as input to a machine learning model to generate a latent space representation of the rotated cropped input image. The system edits the latent space representation and provides the edited latent space representation to a generator neural network to generate a generated edited image. The system applies an inverse affine transformation to rotate the generated edited image and aligns an identified segment of the rotated generated edited image with an identified corresponding segment of the input image to produce an aligned rotated generated edited image. The system blends the aligned rotated generated edited image with the input image to generate an edited output image.

Abstract: Techniques are disclosed for filling or otherwise replacing a target region of a primary image with a corresponding region of an auxiliary image. The filling or replacing can be done with an overlay (no subtractive process need be run on the primary image). Because the primary and auxiliary images may not be aligned, both geometric and photometric transformations are applied to the primary and/or auxiliary images. For instance, a geometric transformation of the auxiliary image is performed, to better align features of the auxiliary image with corresponding features of the primary image. Also, a photometric transformation of the auxiliary image is performed, to better match color of one or more pixels of the auxiliary image with color of corresponding one or more pixels of the primary image. The corresponding region of the transformed auxiliary image is then copied and overlaid on the target region of the primary image.

Abstract: Embodiments of the present invention provide systems, methods, and computer storage media for automatically synthesizing a content-aware fill using similarity transformed patches. A user interface receives a user-specified hole and a user-specified sampling region, both of which may be stored in a constraint mask. A brush tool can be used to interactively brush the sampling region and modify the constraint mask. The mask is passed to a patch-based synthesizer configured to synthesize the fill using similarity transformed patches sampled from the sampling region. Fill properties such as similarity transform parameters can be set to control the manner in which the fill is synthesized. A live preview can be provided with gradual updates of the synthesized fill prior to completion. Once a fill has been synthesized, the user interface presents the original image, replacing the hole with the synthesized fill.

Abstract: Techniques are disclosed for 3D object reconstruction using photometric mesh representations. A decoder is pretrained to transform points sampled from 2D patches of representative objects into 3D polygonal meshes. An image frame of the object is fed into an encoder to get an initial latent code vector. For each frame and camera pair from the sequence, a polygonal mesh is rendered at the given viewpoints. The mesh is optimized by creating a virtual viewpoint, rasterized to obtain a depth map. The 3D mesh projections are aligned by projecting the coordinates corresponding to the polygonal face vertices of the rasterized mesh to both selected viewpoints. The photometric error is determined from RGB pixel intensities sampled from both frames. Gradients from the photometric error are backpropagated into the vertices of the assigned polygonal indices by relating the barycentric coordinates of each image to update the latent code vector.

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 for generating a modified digital image from extracted spatial and global codes. For example, the disclosed systems can utilize a global and spatial autoencoder to extract spatial codes and global codes from digital images. The disclosed systems can further utilize the global and spatial autoencoder to generate a modified digital image by combining extracted spatial and global codes in various ways for various applications such as style swapping, style blending, and attribute editing.

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: Systems and techniques for automatic digital parameter adjustment are described that leverage insights learned from an image set to automatically predict parameter values for an input item of digital visual content. To do so, the automatic digital parameter adjustment techniques described herein captures visual and contextual features of digital visual content to determine balanced visual output in a range of visual scenes and settings. The visual and contextual features of digital visual content are used to train a parameter adjustment model through machine learning techniques that captures feature patterns and interactions. The parameter adjustment model exploits these feature interactions to determine visually pleasing parameter values for an input item of digital visual content. The predicted parameter values are output, allowing further adjustment to the parameter values.

Abstract: Methods and systems are provided for accurately filling holes, regions, and/or portions of images using iterative image inpainting. In particular, iterative inpainting utilize a confidence analysis of predicted pixels determined during the iterations of inpainting. For instance, a confidence analysis can provide information that can be used as feedback to progressively fill undefined pixels that comprise the holes, regions, and/or portions of an image where information for those respective pixels is not known. To allow for accurate image inpainting, one or more neural networks can be used. For instance, a coarse result neural network (e.g., a GAN comprised of a generator and a discriminator) and a fine result neural network (e.g., a GAN comprised of a generator and two discriminators).

Abstract: Methods and systems are provided for accurately filling holes, regions, and/or portions of high-resolution images using guided upsampling during image inpainting. For instance, an image inpainting system can apply guided upsampling to an inpainted image result to enable generation of a high-resolution inpainting result from a lower-resolution image that has undergone inpainting. To allow for guided upsampling during image inpainting, one or more neural networks can be used. For instance, a low-resolution result neural network (e.g., comprised of an encoder and a decoder) and a high-resolution input neural network (e.g., comprised of an encoder and a decoder). The image inpainting system can use such networks to generate a high-resolution inpainting image result that fills the hole, region, and/or portion of the image.