Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for augmenting the functionality of large language models using a hybrid causal-bidirectional attention method. In particular, the disclosed systems generate, from a plurality of tokens interpretable by a large language model, a set of context tokens comprising tokens with bidirectional attention and a set of span tokens comprising tokens with causal attention and bidirectional attention. Additionally, the disclosed systems modify parameters of the large language model at a first training stage by utilizing a first loss function that incorporates the set of context tokens and a second loss function that incorporates the set of span tokens. Further, the disclosed systems modify the parameters of the large language model at a second training stage by utilizing the first loss function, the second loss function, and a third loss function that incorporates the set of context tokens.

Abstract: Embodiments are disclosed for training a system to generate audio and video representations using self-supervised learning. The method may include receiving a video signal including an audio component and a video component. A first machine learning model is trained to determine a representation of the audio component using a contrastive learning task and a temporal learning task. A second machine learning model to determine a representation of the video component using the contrastive learning task and the temporal learning task. By training the machine learning models using both contrastive learning tasks and temporal learning tasks, the machine learning models learn short term features, long term features, and semantic features of input data.

Abstract: A method, apparatus, non-transitory computer readable medium, and system for image generation includes obtaining an input prompt including a first image element and a second image element. The image generation model generates first image features representing the first image element using a first layer selected based on the first image element and second image features representing the second image element using a second layer selected based on the second image element, wherein the second layer is selected based on the second image element. A synthetic image is generated including the first image element and the second image element based on the first image features and the second image features.

Abstract: Digital image visual similarity determination techniques are described. In implementations, a search result is generated based on visual similarity of a plurality of digital images with respect to an input digital image. The search result is generated by locating a plurality of candidate digital images from the plurality of digital images based on a search, calculating spatial feature maps for the input digital image and the plurality of candidate digital images using respective layers of one or more neural networks, and forming a plurality of similarity scores by comparing the spatial feature maps from the plurality of candidate digital images, respectively, with the spatial feature maps for the input digital image.

Abstract: Methods, systems, and non-transitory computer readable storage media are disclosed for generating segmentations of a raster image via a half-edge mesh structure with scanline operations. The disclosed system determines, during scanline operations on a raster image, a plurality of sets of adjacent pixels having a common color value in the raster image. The disclosed system determines, during the scanline operations on the raster image, a plurality of half-edges at edges of pixels along a boundary of a set of adjacent pixels of the plurality of sets of adjacent pixels with next half-edge directions indicating directions of subsequent half-edges along the boundary of the set of adjacent pixels. The disclosed system generates one or more oriented polyline boundary loops representing the boundary of the set of adjacent pixels from the plurality of half-edges and the next half-edge directions of the set of adjacent pixels.

Abstract: Embodiments are disclosed for retrieving videos for a semantic and temporal alignment between a pair of video clips. The method may include receiving a query video clip. The method may further include determining alignment ratios between the query video clip and one or more candidate video clips. The method may further include identifying an alignable video clip from the one or more candidate video clips based on the alignment ratios. The method may further include aligning the alignable video clip with the query video clip.

Abstract: Embodiments are disclosed for retrieving videos for a semantic and temporal alignment between a pair of video clips. The method may include receiving a query video clip. The method may further include determining alignment ratios between the query video clip and one or more candidate video clips. The method may further include identifying an alignable video clip from the one or more candidate video clips based on the alignment ratios. The method may further include aligning the alignable video clip with the query video clip.

Abstract: Digital image visual aesthetic score generation techniques are described. In one or more examples, these techniques are implemented by a system including a training data collection module implemented by a processing device to collect training data including training digital images and user interaction data describing user interaction with the training digital images, respectively. A training module is configured to train a machine-learning model using the training data to generate an aesthetic score based on an input digital image. The aesthetic score is configured to specify an amount of visual aesthetics exhibited by the input digital image.

Abstract: Embodiments are disclosed for performing content authentication. A method of content authentication may include dividing a query video into a plurality of chunks. A feature vector may be generated, using a fingerprinting model, for each chunk from the plurality of chunks. Similar video chunks are identified from a trusted chunk database based on the feature vectors using a multi-chunk search policy. One or more original videos corresponding to the query video are then returned.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for training and implementing a vision-language model using masked distillation and contrastive image-text training. In particular, in one or more embodiments, the disclosed systems generate, utilizing a vision encoder, an image embedding from a masked digital image comprising a digital image with one or more masked patches. In some embodiments, the disclosed systems generate, utilizing a text encoder, a text embedding from a masked text phrase. In one or more embodiments, the disclosed systems generate, utilizing the vision-language model from the image embedding and the text embedding, a predicted text reconstruction of the text description and a predicted image reconstruction of the digital image.

Abstract: Embodiments are disclosed for performing content authentication. A method of content authentication may include dividing a query video into a plurality of chunks. A feature vector may be generated, using a fingerprinting model, for each chunk from the plurality of chunks. Similar video chunks are identified from a trusted chunk database based on the feature vectors using a multi-chunk search policy. One or more original videos corresponding to the query video are then returned.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media that utilize deep learning to map query videos to known videos so as to identify a provenance of the query video or identify editorial manipulations of the query video relative to a known video. For example, the video comparison system includes a deep video comparator model that generates and compares visual and audio descriptors utilizing codewords and an inverse index. The deep video comparator model is robust and ignores discrepancies due to benign transformations that commonly occur during electronic video distribution.

Abstract: Techniques for learning to personalize vision-language models through meta-personalization are described. In one embodiment, one or more processing devices lock a pre-trained vision-language model (VLM) during a training phase. The processing devices train the pre-trained VLM to augment a text encoder of the pre-trained VLM with a set of general named video instances to form a meta-personalized VLM, the meta-personalized VLM to include global category features. The processing devices test the meta-personalized VLM to adapt the text encoder with a set of personal named video instances to form a personal VLM, the personal VLM comprising the global category features personalized with a set of personal instance weights to form a personal instance token associated with the user. Other embodiments are described and claimed.

Abstract: This disclosure describes one or more implementations of systems, non-transitory computer-readable media, and methods that generate a temporally remapped video that satisfies a desired target duration while preserving natural video dynamics. In certain instances, the disclosed systems utilize a playback speed prediction machine-learning model that recognizes and localizes temporally varying changes in video playback speed to re-time a digital video with varying frame-change speeds. For instance, to re-time the digital video, the disclosed systems utilize the playback speed prediction machine-learning model to infer the slowness of individual video frames. Subsequently, in certain embodiments, the disclosed systems determine, from frames of a digital video, a temporal frame sub-sampling that is consistent with the slowness predictions and fit within a target video duration.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media that utilize deep learning to map query videos to known videos so as to identify a provenance of the query video or identify editorial manipulations of the query video relative to a known video. For example, the video comparison system includes a deep video comparator model that generates and compares visual and audio descriptors utilizing codewords and an inverse index. The deep video comparator model is robust and ignores discrepancies due to benign transformations that commonly occur during electronic video distribution.

Abstract: Embodiments are disclosed for identifying matching content using neural content fingerprints. The method may include receiving a request to identify content matching a query content item, wherein the query content item is a time varying content item, generating, by an embedding network, a neural fingerprint for the query content item, identifying one or more candidate content items based on the neural fingerprint of the query content item, determining, by a ranking network, one or more similarity scores corresponding to the one or more candidate content items, and identifying one or more matching content items based on the one or more similarity scores.

Abstract: The disclosed invention includes systems and methods for training and employing equivariant models for generating representations (e.g., vector representations) of temporally-varying content, such as but not limited to video content. The trained models are equivariant to temporal transformations applied to the input content (e.g., video content). The trained models are additionally invariant to non-temporal transformations (e.g., spatial and/or color-space transformations) applied to the input content. Such representations are employed in various machine learning tasks, such as but not limited to video retrieval (e.g., video search engine applications), identification of actions depicted in video, and temporally ordering clips of the video.

Abstract: Embodiments are disclosed for training a system to generate audio and video representations using self-supervised learning. The method may include receiving a video signal including an audio component and a video component. A first machine learning model is trained to determine a representation of the audio component using a contrastive learning task and a temporal learning task. A second machine learning model to determine a representation of the video component using the contrastive learning task and the temporal learning task. By training the machine learning models using both contrastive learning tasks and temporal learning tasks, the machine learning models learn short term features, long term features, and semantic features of input data.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media that utilize deep learning to map query videos to known videos so as to identify a provenance of the query video or identify editorial manipulations of the query video relative to a known video. For example, the video comparison system includes a deep video comparator model that generates and compares visual and audio descriptors utilizing codewords and an inverse index. The deep video comparator model is robust and ignores discrepancies due to benign transformations that commonly occur during electronic video distribution.

Abstract: This disclosure describes one or more implementations of systems, non-transitory computer-readable media, and methods that generate a temporally remapped video that satisfies a desired target duration while preserving natural video dynamics. In certain instances, the disclosed systems utilize a playback speed prediction machine-learning model that recognizes and localizes temporally varying changes in video playback speed to re-time a digital video with varying frame-change speeds. For instance, to re-time the digital video, the disclosed systems utilize the playback speed prediction machine-learning model to infer the slowness of individual video frames. Subsequently, in certain embodiments, the disclosed systems determine, from frames of a digital video, a temporal frame sub-sampling that is consistent with the slowness predictions and fit within a target video duration.