Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for generating semantic scene graphs for digital images using an external knowledgebase for feature refinement. For example, the disclosed system can determine object proposals and subgraph proposals for a digital image to indicate candidate relationships between objects in the digital image. The disclosed system can then extract relationships from an external knowledgebase for refining features of the object proposals and the subgraph proposals. Additionally, the disclosed system can generate a semantic scene graph for the digital image based on the refined features of the object/subgraph proposals. Furthermore, the disclosed system can update/train a semantic scene graph generation network based on the generated semantic scene graph. The disclosed system can also reconstruct the image using object labels based on the refined features to further update/train the semantic scene graph generation network.

Abstract: In order to provide monocular depth prediction, a trained neural network may be used. To train the neural network, edge detection on a digital image may be performed to determine at least one edge of the digital image, and then a first point and a second point of the digital image may be sampled, based on the at least one edge. A relative depth between the first point and the second point may be predicted, and the neural network may be trained to perform monocular depth prediction using a loss function that compares the predicted relative depth with a ground truth relative depth between the first point and the second point.

Abstract: The present disclosure is directed toward systems and methods for detecting an object in an input image based on a target object keyword. For example, one or more embodiments described herein generate a heat map of the input image based on the target object keyword and generate various bounding boxes based on a pixel analysis of the heat map. One or more embodiments described herein then utilize the various bounding boxes to determine scores for generated object location proposals in order to provide a highest scoring object location proposal overlaid on the input image.

Abstract: Systems and methods provide editing operations in a smart editing system that may generate a focal point within a mask of an object for each frame of a video segment and perform editing effects on the frames of the video segment to quickly provide users with natural video editing effects. An eye-gaze network may produce a hotspot map of predicted focal points in a video frame. These predicted focal points may then be used by a gaze-to-mask network to determine objects in the image and generate an object mask for each of the detected objects. This process may then be repeated to effectively track the trajectory of objects and object focal points in videos. Based on the determined trajectory of an object in a video clip and editing parameters, the editing engine may produce editing effects relative to an object for the video clip.

Abstract: A Video Semantic Segmentation System (VSSS) is disclosed that performs accurate and fast semantic segmentation of videos using a set of temporally distributed neural networks. The VSSS receives as input a video signal comprising a contiguous sequence of temporally-related video frames. The VSSS extracts features from the video frames in the contiguous sequence and based upon the extracted features, selects, from a set of labels, a label to be associated with each pixel of each video frame in the video signal. In certain embodiments, a set of multiple neural networks are used to extract the features to be used for video segmentation and the extraction of features is distributed among the multiple neural networks in the set. A strong feature representation representing the entirety of the features is produced for each video frame in the sequence of video frames by aggregating the output features extracted by the multiple neural networks.

Abstract: A query image is received, along with a query to initiate a search process to find other images based on the query image. The query includes a preference value associated with an attribute, the preference value indicative of a level of emphasis to be placed on the attribute during the search. A full query vector, which is within a first dimensional space and representative of the query image, is generated. The full query vector is projected to a reduced dimensional space having a dimensionality lower than the first dimensional space, to generate a query vector. An attribute direction corresponding to the attribute is identified. A plurality of candidate vectors of the reduced dimensional space is searched, based on the attribute direction, the query vector, and the preference value, to identify a target vector of the plurality of candidate vectors. A target image, representative of the target vector, is displayed.

Abstract: A non-transitory computer-readable medium includes program code that is stored thereon. The program code is executable by one or more processing devices for performing operations including generating, by a model that includes trainable components, a learned image representation of a target image. The operations further include generating, by a text embedding model, a text embedding of a text query. The text embedding and the learned image representation of the target image are in a same embedding space. Additionally, the operations include generating a class activation map of the target image by, at least, convolving the learned image representation of the target image with the text embedding of the text query. Moreover, the operations include generating an object-segmented image using the class activation map of the target image.

Abstract: In implementations of object detection in images, object detectors are trained using heterogeneous training datasets. A first training dataset is used to train an image tagging network to determine an attention map of an input image for a target concept. A second training dataset is used to train a conditional detection network that accepts as conditional inputs the attention map and a word embedding of the target concept. Despite the conditional detection network being trained with a training dataset having a small number of seen classes (e.g., classes in a training dataset), it generalizes to novel, unseen classes by concept conditioning, since the target concept propagates through the conditional detection network via the conditional inputs, thus influencing classification and region proposal. Hence, classes of objects that can be detected are expanded, without the need to scale training databases to include additional classes.

Abstract: Digital image completion by learning generation and patch matching jointly is described. Initially, a digital image having at least one hole is received. This holey digital image is provided as input to an image completer formed with a dual-stage framework that combines a coarse image neural network and an image refinement network. The coarse image neural network generates a coarse prediction of imagery for filling the holes of the holey digital image. The image refinement network receives the coarse prediction as input, refines the coarse prediction, and outputs a filled digital image having refined imagery that fills these holes. The image refinement network generates refined imagery using a patch matching technique, which includes leveraging information corresponding to patches of known pixels for filtering patches generated based on the coarse prediction. Based on this, the image completer outputs the filled digital image with the refined imagery.

Abstract: The present disclosure relates to systems, method, and computer readable media that iteratively apply a neural network to a digital image at a reduced resolution to automatically identify pixels of salient objects portrayed within the digital image. For example, the disclosed systems can generate a reduced-resolution digital image from an input digital image and apply a neural network to identify a region corresponding to a salient object. The disclosed systems can then iteratively apply the neural network to additional reduced-resolution digital images (based on the identified region) to generate one or more reduced-resolution segmentation maps that roughly indicate pixels of the salient object. In addition, the systems described herein can perform post-processing based on the reduced-resolution segmentation map(s) and the input digital image to accurately determine pixels that correspond to the salient object.

Abstract: Systems and methods for natural language processing are described. Embodiments are configured to receive a structured representation of a search query, wherein the structured representation comprises a plurality of nodes and at least one edge connecting two of the nodes, receive a modification expression for the search query, wherein the modification expression comprises a natural language expression, generate a modified structured representation based on the structured representation and the modification expression using a neural network configured to combine structured representation features and natural language expression features, and perform a search based on the modified structured representation.

Abstract: In some embodiments, an image manipulation application receives an incomplete image that includes a hole area lacking image content. The image manipulation application applies a contour detection operation to the incomplete image to detect an incomplete contour of a foreground object in the incomplete image. The hole area prevents the contour detection operation from detecting a completed contour of the foreground object. The image manipulation application further applies a contour completion model to the incomplete contour and the incomplete image to generate the completed contour for the foreground object. Based on the completed contour and the incomplete image, the image manipulation application generates image content for the hole area to generate a completed image.

Abstract: A distractor detector includes a heatmap network and a distractor classifier. The heatmap network operates on an input image to generate a heatmap for a main subject, a heatmap for a distractor, and optionally a heatmap for the background. Each object is cropped within the input image to generate a corresponding cropped image. Regions within the heatmaps that correspond to the objects are identified, and each of the regions is cropped within each of the heatmaps to generate cropped heatmaps. The distractor classifier then operates on the cropped images and the cropped heatmaps to classify each of the objects as being either a main subject or a distractor.

Abstract: The present disclosure discloses an electromagnetic regulating valve with a check function. The valve may effectively regulate the flow in a valve body. Due to different lengths and taper angles of a valve flap group, the flow can be regulated more finely, and a medium can be prevented from backflow by a check function when necessary. Meanwhile, a mesh structure of bases of valve flaps can effectively reduce the instability of flow field motion so as to achieve an effect of rectifying.

Abstract: In implementations of multi-resolution color-based image search, an image search system determines a color vector for a query image based on a color histogram of the query image by concatenating two color histograms having different resolutions. The image search system can compute distance measures between the color vector of the query image and color vectors of candidate images. The image search system can select one or more of the candidate images to return based on the distance measures utilizing the distance measures as indication of color similarity of the candidate images to the query image.

Abstract: The present disclosure relates to methods, systems, and non-transitory computer-readable media for retrieving digital images in response to queries. For example, in one or more embodiments, the disclosed systems receive a query comprising text and generates a cross-lingual-multimodal embedding for the text within a multimodal embedding space. The disclosed systems further identifies an image embedding for a digital image that corresponds to (e.g., is relevant to) the text from the query based on an embedding distance between the image embedding and the cross-lingual-multimodal embedding for the text within the multimodal embedding space. Accordingly, the disclosed systems retrieve the digital image associated with the image embedding for display on a client device, such as the client device that submitted the query.

Abstract: The present disclosure relates to a color classification system that accurately classifies objects in digital images based on color. In particular, in one or more embodiments, the color classification system utilizes a multidimensional color space and one or more color mappings to match objects to colors. Indeed, the color classification system can accurately and efficiently detect the color of an object utilizing one or more color similarity regions generated in the multidimensional color space.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for accurately and efficiently learning parameters of a distilled neural network from parameters of a source neural network utilizing multiple augmentation strategies. For example, the disclosed systems can generate lightly augmented digital images and heavily augmented digital images. The disclosed systems can further learn parameters for a source neural network from the lightly augmented digital images. Moreover, the disclosed systems can learn parameters for a distilled neural network from the parameters learned for the source neural network. For example, the disclosed systems can compare classifications of heavily augmented digital images generated by the source neural network and the distilled neural network to transfer learned parameters from the source neural network to the distilled neural network via a knowledge distillation loss function.

Abstract: Enhanced methods and systems for the semantic segmentation of images are described. A refined segmentation mask for a specified object visually depicted in a source image is generated based on a coarse and/or raw segmentation mask. The refined segmentation mask is generated via a refinement process applied to the coarse segmentation mask. The refinement process correct at least a portion of both type I and type II errors, as well as refine boundaries of the specified object, associated with the coarse segmentation mask. Thus, the refined segmentation mask provides a more accurate segmentation of the object than the coarse segmentation mask. A segmentation refinement model is employed to generate the refined segmentation mask based on the coarse segmentation mask. That is, the segmentation model is employed to refine the coarse segmentation mask to generate more accurate segmentations of the object. The refinement process is an iterative refinement process carried out via a trained neural network.

Abstract: Systems, methods, and non-transitory computer-readable media are disclosed for determining multi-term contextual tags for digital content and propagating the multi-term contextual tags to additional digital content. For instance, the disclosed systems can utilize search query supervision to determine and associate multi-term contextual tags (e.g., tags that represent a specific concept based on the order of the terms in the tag) with digital content. Furthermore, the disclosed systems can propagate the multi-term contextual tags determined for the digital content to additional digital content based on similarities between the digital content and additional digital content (e.g., utilizing clustering techniques). Additionally, the disclosed systems can provide digital content as search results based on the associated multi-term contextual tags.