Facebook, Inc.

Benjamin Ray

Balmanohar Paluri

Sofus Attila Macskassy

Ahmad Abdulmageed Mohammed Abdulkader

Du Le Hong Tran

Karthik Subbian

Anurag Ranjan

In one embodiment, a method includes, by a computing device, receiving a plurality of inputs for a convolution layer of a convolutional neural network, the convolution layer having one or more input channels and one or more output channels, wherein the inputs are received via the input channels, generating, by convolving the inputs with one or more two-dimensional filters, a plurality of intermediate values, and generating, by convolving the intermediate values with one or more one-dimensional filters, a plurality of outputs, wherein the one-dimensional filters receive the intermediate values from the two-dimensional filters via intermediate channels. The method may provide the outputs to a subsequent layer of the convolutional neural network via the output channels. Each of the two dimensions of the two-dimensional filter may correspond to a spatial dimension, and the one dimension of the one-dimensional filter may correspond to a temporal dimension.

Video analysis using convolutional networks

Systems, methods, and non-transitory computer-readable media can acquire an image that depicts at least one character. A set of pixels, within the image, through which the at least one character is depicted can be identified. At least one linear portion, within the image, can be identified based on the set of pixels. For each sub-portion within the at least one linear portion, a respective first confidence score representing a respective first likelihood that a respective sub-portion depicts the at least one character can be determined.

Systems and methods for incremental character recognition to recognize characters in images

In one embodiment, a method includes a computing system accessing a first training data comprising a first image and a second image and an associated optical flow estimation. The system may input (1) the first image into a first machine-learning model configured to generate a first output and (2) the optical flow estimation into a second machine-learning model configured to generate a second output. The first output of the first machine-learning model is associated with first image segments of a predetermined number, and the second output of the second machine-learning model is associated with transformations of the predetermined number. The first output, the transformations, and the first image are configured to generate an estimated image. The system trains the first machine-learning model and the second machine-learning model based on at least a comparison of the estimated image and the second image.

Unsupervised video segmentation

Systems, methods, and non-transitory computer-readable media can identify a media content item for which media processing is to be performed. State information associated with the media content item can be acquired. At least some of the media processing can be enabled, based on the state information, to be performed client-side with respect to the media content item. The state information can indicate a next processing step of the at least some of the media processing that is to be performed. The state information can be updated based on the at least some of the media processing performed client-side.

Growing community of inventors

Palo Alto, CA, United States of America

Benjamin Ray