Abstract: In one embodiment, a method includes receiving a plurality of search queries comprising n-grams; identifying a subset of the plurality of search queries as being queries for visual-media items based on one or more n-grams of the search query being associated with visual-media content; calculating, for each of the n-grams of the search queries of the subset, a popularity-score based on a count of the search queries in the subset that include the n-gram; determining popular n-grams, wherein each of the popular n-grams is an n-gram of the search queries of the subset of search queries having a popularity-score greater than a threshold popularity-score; and selecting one or more of the popular n-grams for training a visual-concept recognition system, wherein each of the popular n-grams is selected based on whether it is associated with a visual concept.

Abstract: In one embodiment, a method includes accessing a first feature vector representing a video-content object corresponding to a node in a social graph, wherein the video-content object comprises frames and audio and is associated with text, the first feature vector is based on one or more of the frames; accessing a second feature vector representing the video-content object, wherein the second feature vector is based on at least some of the text; accessing a third feature vector representing the video-content object, wherein the third feature vector is based on one or more portions of the audio; determining a fourth feature vector representing the video-content object, wherein the fourth feature vector is based on a combination of the first, second, and third feature vectors; and determining a context of the video-content object based on the fourth feature vector and social-graph information.

Abstract: In one embodiment, a method includes identifying a shared visual concept in visual-media items based on shared visual features in images of the visual-media items; extracting, for each of the visual-media items, n-grams from communications associated with the visual-media item; generating, in a d-dimensional space, an embedding for each of the visual-media items at a location based on the visual concepts included in the visual-media item; generating, in the d-dimensional space, an embedding for each of the extracted n-grams at a location based on a frequency of occurrence of the n-gram in the communications associated with the visual-media items; and associating, with the shared visual concept, the extracted n-grams that have embeddings within a threshold area of the embeddings for the identified visual-media items.

Abstract: In one embodiment, a method includes detecting one or more objects in an image, generating at least one mask for each of the detected objects, wherein each of the masks is defined by a perimeter, classifying the detected objects, receiving gesture input in relation to the image, determining whether one or more locations associated with the gesture input correlate with any of the masks, and providing feedback regarding the image in response to the gesture input. Each of the masks may include data identifying the corresponding detected object, and the perimeter of each mask may correspond to a perimeter of the corresponding detected object. The perimeter of the corresponding detected object may separate the detected object from one or more portions of the image that are distinct from the detected object.

Abstract: In one embodiment, a system retrieves a first feature vector for an image. The image is inputted into a first deep-learning model, which is a first-version model, and the first feature vector may be output from a processing layer of the first deep-learning model for the image. The first feature vector using a feature-vector conversion model to obtain a second feature vector for the image. The feature-vector conversion model is trained to convert first-version feature vectors to second-version feature vectors. The second feature vector is associated with a second deep-learning model, and the second deep-learning model is a second-version model. The second-version model is an updated version of the first-version model. A plurality of predictions for the image may be generated using the second feature vector and the second deep-learning model.

Abstract: In one embodiment, a method includes accessing a first feature vector representing a video-content object corresponding to a node in a social graph, wherein the video-content object comprises frames and audio and is associated with text, the first feature vector is based on one or more of the frames; accessing a second feature vector representing the video-content object, wherein the second feature vector is based on at least some of the text; accessing a third feature vector representing the video-content object, wherein the third feature vector is based on one or more portions of the audio; determining a fourth feature vector representing the video-content object, wherein the fourth feature vector is based on a combination of the first, second, and third feature vectors; and determining a context of the video-content object based on the fourth feature vector and social-graph information.

Abstract: In one embodiment, a method includes accessing for each of a plurality of input images a feature vector corresponding to the input image and metadata indicating a relationship of the input image to a predetermined concept; training with machine learning an image classifier associated with the predetermined concept based on the feature vectors of the input images and the metadata indicating their respective relationships to the predetermined concept; accessing for each of a plurality of evaluation images a feature vector that corresponds to the evaluation image; for each evaluation image calculating with the image classifier as trained a score indicating how closely related the evaluation image is to the predetermined concept, based on the feature vector corresponding to the evaluation image; and providing for display to a user one or more of the evaluation images and their respective scores as calculated by the image classifier.

Abstract: In one embodiment an image is processed using a deep-learning model to obtain one or more first predictions. Each first prediction is a likelihood that a respective first concept in a set of first concepts is associated with the image. A feature vector is retrieved for the image. The feature vector is output from a processing layer of the deep-learning model for the image. The feature vector for the image is processed using a first linear model, where the first linear model is trained to detect one or more second concepts. One or more second predictions are obtained from the first linear model. Each second prediction is a likelihood that a respective second concept of the one or more second concepts is associated with the image.

Abstract: In one embodiment, a method includes receiving a request for a protected resource, providing information to display a challenge-response test, where the challenge-response test includes an image and instructions to provide user input in relation to the image, the image comprises one or more masks, and each of the masks is defined by a perimeter, receiving user input in relation to the image, generating an assessment of the user input based on a correlation between the user input and the masks, determining, based on the assessment, whether the user input corresponds to human-generated input, and if the user input may be deemed responsive to the instructions, then providing information to access the protected resource, else providing information indicating that the user input failed the challenge-response test. Each of the masks may include a classification, and the instructions may provide user input in relation to the classifications.

Abstract: In one embodiment, a method includes receiving a query of a first user; retrieving videos that match the query; determining a filtered set of videos, wherein the filtering includes removing duplicate videos based on the duplicate videos having a digital fingerprint that is within a threshold degree of sameness from that of a modal video; calculating, for each video, similarity-scores that correspond to a degree of similarity between the video and another video in the filtered set; grouping the videos into clusters that include videos with similarity-scores greater than a threshold similarity-score with respect to each other video in the cluster; and sending, to the first user, a search-results interface including search results for the videos that are organized within the interface based on the respective clusters of their corresponding videos.

Abstract: In one embodiment, a method includes receiving a plurality of search queries comprising n-grams; identifying a subset of the plurality of search queries as being queries for visual-media items based on one or more n-grams of the search query being associated with visual-media content; calculating, for each of the n-grams of the search queries of the subset, a popularity-score based on a count of the search queries in the subset that include the n-gram; determining popular n-grams, wherein each of the popular n-grams is an n-gram of the search queries of the subset of search queries having a popularity-score greater than a threshold popularity-score; and selecting one or more of the popular n-grams for training a visual-concept recognition system, wherein each of the popular n-grams is selected based on whether it is associated with a visual concept.

Abstract: In one embodiment, a method includes identifying a shared visual concept in visual-media items based on shared visual features in images of the visual-media items; extracting, for each of the visual-media items, n-grams from communications associated with the visual-media item; generating, in a d-dimensional space, an embedding for each of the visual-media items at a location based on the visual concepts included in the visual-media item; generating, in the d-dimensional space, an embedding for each of the extracted n-grams at a location based on a frequency of occurrence of the n-gram in the communications associated with the visual-media items; and associating, with the shared visual concept, the extracted n-grams that have embeddings within a threshold area of the embeddings for the identified visual-media items.

Abstract: In one embodiment, a method includes receiving a query from a user for videos; identifying videos matching the query; retrieving, for each identified video, a set of keyframes that are associated with one or more concepts; calculating, for each keyframe of each identified video, a keyframe-score based on a prevalence of the concepts associated with the keyframe, determined with reference to the concepts associated with each other keyframe in the set of retrieved keyframes for the identified video; and sending, to the first user, a search-results interface including search results corresponding to one or more of the identified videos, each search result comprising keyframes for the corresponding identified video having keyframe-scores greater than a threshold keyframe-score.

Abstract: In one embodiment, a method may include receiving a first content item. A first embedding of the first content item may be determined and may corresponds to a first point in an embedding space. The embedding space may include a plurality of second points corresponding to a plurality of second embeddings of second content items. The embeddings are determined using a deep-learning model. The points are located in one or more clusters in the embedding space, which are each associated with a class of content items. Locations of points within clusters may be based on one or more attributes of the respective corresponding content items. Second content items that are similar to the first content item may be identified based on the locations of the first point and the second points and on particular clusters that the second points corresponding to the identified second content items are located in.

Abstract: The techniques introduced here include a system and method for transcoding multimedia content based on the results of content analysis. The determination of specific transcoding parameters, used for transcoding multimedia content, can be performed by utilizing the results of content analysis of the multimedia content. One of the results of the content analysis is the determination of image type of any images included in the multimedia content. The content analysis uses one or more of several techniques, including analyzing content metadata, examining colors of contiguous pixels in the content, using histogram analysis, using compression distortion analysis, analyzing image edges, or examining user provided inputs. Transcoding the multimedia content can include adapting the content to the constraints in delivery and display, processing and storage of user computing devices.

Abstract: The techniques introduced here include a system and method for transcoding multimedia content based on the results of content analysis. The determination of specific transcoding parameters, used for transcoding multimedia content, can be performed by utilizing the results of content analysis of the multimedia content. One of the results of the content analysis is the determination of image type of any images included in the multimedia content. The content analysis uses one or more of several techniques, including analyzing content metadata, examining colors of contiguous pixels in the content, using histogram analysis, using compression distortion analysis, analyzing image edges, or examining user provided inputs. Transcoding the multimedia content can include adapting the content to the constraints in delivery and display, processing and storage of user computing devices.

Abstract: The techniques introduced here include a system and method for transcoding multimedia content based on the results of content analysis. The determination of specific transcoding parameters, used for transcoding multimedia content, can be performed by utilizing the results of content analysis of the multimedia content. One of the results of the content analysis is the determination of image type of any images included in the multimedia content. The content analysis uses one or more of several techniques, including analyzing content metadata, examining colors of contiguous pixels in the content, using histogram analysis, using compression distortion analysis, analyzing image edges, or examining user provided inputs. Transcoding the multimedia content can include adapting the content to the constraints in delivery and display, processing and storage of user computing devices.

Abstract: The techniques introduced here include a system and method for transcoding multimedia content based on the results of content analysis. The determination of specific transcoding parameters, used for transcoding multimedia content, can be performed by utilizing the results of content analysis of the multimedia content. One of the results of the content analysis is the determination of image type of any images included in the multimedia content. The content analysis uses one or more of several techniques, including analyzing content metadata, examining colors of contiguous pixels in the content, using histogram analysis, using compression distortion analysis, analyzing image edges, or examining user provided inputs. Transcoding the multimedia content can include adapting the content to the constraints in delivery and display, processing and storage of user computing devices.

Abstract: The techniques introduced here include a system and method for transcoding multimedia content based on the results of content analysis. The determination of specific transcoding parameters, used for transcoding multimedia content, can be performed by utilizing the results of content analysis of the multimedia content. One of the results of the content analysis is the determination of image type of any images included in the multimedia content. The content analysis uses one or more of several techniques, including analyzing content metadata, examining colors of contiguous pixels in the content, using histogram analysis, using compression distortion analysis, analyzing image edges, or examining user provided inputs. Transcoding the multimedia content can include adapting the content to the constraints in delivery and display, processing and storage of user computing devices.

Abstract: The techniques introduced here include a system and method for transcoding multimedia content based on the results of content analysis. The determination of specific transcoding parameters, used for transcoding multimedia content, can be performed by utilizing the results of content analysis of the multimedia content. One of the results of the content analysis is the determination of image type of any images included in the multimedia content. The content analysis uses one or more of several techniques, including analyzing content metadata, examining colors of contiguous pixels in the content, using histogram analysis, using compression distortion analysis, analyzing image edges, or examining user provided inputs. Transcoding the multimedia content can include adapting the content to the constraints in delivery and display, processing and storage of user computing devices.