Abstract: A processor-implemented method for processing a video includes receiving the video as an input at an artificial neural network (ANN). The video includes a sequence of frames. A set of features of a current frame of the video and a prior frame of the video are extracted. The set of features including a set of support features for a set of pixels of the prior frame to be aligned with a set of reference features of the current frame. A similarity between a support feature for each pixel in the set of pixels of the set of support features of the prior frame and a corresponding reference feature of the current frame is computed. An attention map is generated based on the similarity. An output including a reconstruction of the current frame is generated based on the attention map.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for processing multidimensional content using neural networks. An example method generally includes decomposing a multidimensional input into a plurality of two-dimensional subspaces, wherein the plurality of two-dimensional subspaces share a common dimension. A first attention matrix is generated based on a projection of tokens in a first two-dimensional subspace of the plurality of two-dimensional subspaces via an attention block of a transformer neural network, and a second attention matrix is generated based on a projection of tokens in a second two-dimensional subspace of the plurality of two-dimensional subspaces via the attention block of the transformer neural network. An output of the transformer neural network is generated based on a combination of the first attention matrix and the second attention matrix.

Abstract: Certain aspects of the present disclosure provide techniques and apparatus for improved attention-based machine learning. A first attention propagation output is generated using a first transformer block of a plurality of transformer blocks, this generation including processing input data for the first transformer block using a first self-attention sub-block of the first transformer block. The first attention propagation output is propagated to a second transformer block of the plurality of transformer blocks. An output for the second transformer block is generated, this generation including generating output features for the second transformer block based on the first attention propagation output.

Abstract: Systems and techniques are provided for performing holistic video understanding. For example a process can include obtaining a first video and determining, using a machine learning model decision engine, a first machine learning model from a set of machine learning models to use for processing at least a portion of the first video. The first machine learning model can be determined based on one or more characteristics of at least the portion of the first video. The process can include processing at least the portion of the first video using the first machine learning model.

Abstract: A processor-implemented method of video processing using includes receiving, via an artificial neural network (ANN), a video including a first frame and a second frame. A saliency map is generated based on the first frame of the video. The second frame of the video is sampled based on the saliency map. A first portion of the second frame is sampled at a first resolution and a second portion of the second frame is sampled at a second resolution. The first resolution is different than the second resolution. A resampled second frame is generated based on the sampling of the second frame. The resampled second frame is processed to determine an inference associated with the video.

Abstract: Systems and techniques are provided for performing holistic video understanding. For example a process can include obtaining a first video and determining, using a machine learning model decision engine, a first machine learning model from a set of machine learning models to use for processing at least a portion of the first video. The first machine learning model can be determined based on one or more characteristics of at least the portion of the first video. The process can include processing at least the portion of the first video using the first machine learning model.

Abstract: A processor-implemented method for processing a video includes receiving the video as an input at an artificial neural network (ANN). The video includes a sequence of frames. A set of features of a current frame of the video and a prior frame of the video are extracted. The set of features including a set of support features for a set of pixels of the prior frame to be aligned with a set of reference features of the current frame. A similarity between a support feature for each pixel in the set of pixels of the set of support features of the prior frame and a corresponding reference feature of the current frame is computed. An attention map is generated based on the similarity. An output including a reconstruction of the current frame is generated based on the attention map.

Abstract: Certain aspects of the present disclosure provide techniques for processing with an auto exiting machine learning model architecture, including processing input data in a first portion of a classification model to generate first intermediate activation data; providing the first intermediate activation data to a first gate; making a determination by the first gate whether or not to exit processing by the classification model; and generating a classification result from one of a plurality of classifiers of the classification model.

Abstract: A method of pixel-wise localization of an actor and an action in a sequence of frames includes receiving a natural language query describing the action and the actor. The method also includes receiving the sequence of frames. The method further includes localizing the action and the actor in the sequence of frames based on the natural language query.

Abstract: A method of pixel-wise localization of an actor and an action in a sequence of frames includes receiving a natural language query describing the action and the actor. The method also includes receiving the sequence of frames. The method further includes localizing the action and the actor in the sequence of frames based on the natural language query.