Facebook, Inc.

Radisys Corporation

Michael Hamilton Coward

Amit Puntambekar

Minchuan Chen

Wurzel David Parsons-Keir

Sonal Gandhi

Melanie Sclar

Shankar Lakshmi Regunathan

Yaming He

Evgeny V Kuzyakov

David Young Joon Pio

Mathieu R Henaire

Manish S Singh

Thomas George

Sabyasachi Roy

Randall Eugene Rizun

Renuka Bhalerao

Various of the disclosed embodiments relate to multiple video encoders that are used to simultaneously encode a video using encoders configured using different encoding parameters. A segment selector selects an encoded version of the encoded video segment using operational criteria such as video quality and bandwidth. A configuration determination module may analyze the video segment to make a decision about which encoding parameter configurations may be suitable for encoding the video segment. The configuration determination module may be trainable, based on past encoding results.

Ultra-high video compression

A method and system for allocating a variable number of bits per frame in a distributed video encoding using a complexity analyzer is disclosed. The method includes receiving an input video for allocating the number of bits for each segment based on the complexity of the segment. Further, the method includes splitting the input video into plurality of segments. Further, the method includes determining the number of bits to be allocated to the plurality of segments based on the complexity measurement of the input video. In an embodiment, single complexity analyzer can be used to determine the complexity of the plurality of segments. In another embodiment, separate complexity analyzer can be used for each segment to determine the complexity. Further, the method includes allocating the bits to the plurality of segments of the input video. Further, the method includes combining the plurality of segments to form a single output video; thereby, obtaining the encoded output video.

Variable bitrate control for distributed video encoding

In one embodiment, a method includes receiving multimedia content information associated with at least one segment of a multimedia content, receiving a request to view the at least one segment of the multimedia content from a client device, logging playback information associated with the viewing of the at least one segment of the multimedia content, determining a multimedia quality metric associated with the at least one segment of multimedia content based in part upon a portion of the received multimedia content information and a portion of the logged playback information, and classifying the at least one segment of the multimedia content with the multimedia quality metric.

Visual quality metrics

A neural network learning algorithm obtains a fingerprint of a video sequence. Using the fingerprint, selection of an appropriate video stabilization algorithm, and its parameter settings, is performed. The video thus stabilized is made more appealing for viewing and easier for subsequent video compression. The neural network may be trained during a training phase to select a correct fingerprint of each video sequence based on the effectiveness of resulting video stabilization and/or compression.

Neural network to optimize video stabilization parameters

Systems, methods, and non-transitory computer-readable media can present one or more base segments of a first stream of a content item in a viewport interface, the content item being composed using a set of streams that each capture at least one scene from a particular direction, wherein the viewport interface is provided through a display screen of the computing device. A determination is made that a direction of the viewport interface has changed to a different direction during playback of a first base segment of the first stream. One or more offset segments of a second stream that correspond to the different direction are presented in the viewport interface, the offset segments being offset from the set of base segments of the first stream.

Systems and methods for providing content

Video encoding operation is performed in a distributed video encoding system to meet a service layer agreement such as a target time for completion of the video encoding operation. Each of multiple videos being encoded is split into smaller jobs corresponding to video segments of the videos. The jobs are processed according to a job queue. The sequence of jobs in the job queue is periodically updated such that jobs corresponding to each video are processed within the corresponding target encoding time.

Distributed control of video encoding speeds

Video foreground and background detection is performed on a received video sequence on a macroblock by macroblock basis using motion vector encoding information associated with each macroblock. Some of the macroblocks in each frame, for example, edge or corner macroblocks, may be exempt from the foreground and background detection. The detection information, along with the video data, is passed to an image stabilization algorithm that prepares the video for subsequent video compression.

Foreground detection for video stabilization

A method and system for encoding a segment of an input video with accurate placement of key frames, e.g., I-Frames, using Group of Pictures (GOPs) length for output in a distributed video encoding system with two or more encoders is disclosed. The method includes creating overlapping segments in order to allow precise key frame placement. Further, the method includes using desired segment length and the desired key frame placement to create the overlapping segments. Further, the method includes skipping a certain number of frames at the beginning of the segment, and a certain number of frames at the end of the segment, thereby achieving accurate placement of key frames.

Key frame placement for distributed video encoding

An online system more efficiently streams multimedia content over the Internet for play back on client devices with varying computing power and network bandwidths by generating enhanced manifest files that more efficiently identify suitable media representations of the multimedia content. Each media representation has multiple media segments according to predefined byte ranges and a manifest file, which identifies location of the media file, bitrates, resolution, byte range, total duration, and other metadata. The online system customizes a manifest file for a user based on various factors including device capacity, network connectivity type and geolocation of the user. The online system also generates manifest fetch commands, which more efficiently fetch media segments for streaming. In response to changes of streaming server and media file (e.g., increased popularity), the online system dynamically updates corresponding manifest files.

Customizing manifest file for enhancing media streaming

Effectively fetch media content for enhancing media streaming

In one embodiment, a computer server machine generates a time map associated with the video stream. The video stream comprises one or more time slices. The computer server machine identifies a non-relevant time frame. The computer machine receives a comment and a localized comment time stamp, and associates the comment with one or more time slices. One or more client devices communicates a request for one or more time slices to the computer server machine, and the computer server machine determines whether the localized comment time stamp is within the requested one or more time slices. When the localized comment time stamp is within the requested one or more time slices, computer server machine transforms the localized time stamp to a synchronized comment time stamp. Computer server machine sends information regarding the synchronized comment time stamp to the one or more client devices.

Comment synchronization in a video stream

Various of the disclosed embodiments relate to a distributed video encoding or transcoding system may utilize multiple encoding nodes to encode a video sequence by splitting the video into multiple smaller video segments. The assignment of video segments to the encoding nodes is performed to balance the use of the encoding nodes by selecting a node based on its encoding capabilities, e.g., whether the node employed a central processing unit (CPU) based encoding or a graphics processor unit (GPU) based encoding.

Using worker nodes in a distributed video encoding system

In a distributed video encoding system, a video is encoded by splitting into video segments and encoding the segments using multiple encoders. Prior to segmenting the video for distributed video encoding, image stabilization is performed on the video. For each frame in the video, a corresponding transform operation is first computed based on an estimated camera movement. Next, the video is segmented into multiple video segments and the corresponding per-frame transform information for the multiple video segments. The video segments are then distributed to multiple processing nodes that perform the image stabilization of the corresponding video segment by applying the corresponding transform. The results from all the stabilized video segments are then stitched back together for further video encoding operation.

Distributed image stabilization

Systems, methods, and non-transitory computer-readable media can generate a first set of fingerprints that correspond to at least one media stream of a content item. A second set of fingerprints that correspond to at least one media stream of an encoded version of the content item are generated. A first curve is generated based at least in part on the first set of fingerprints. A second curve is generated based at least in part on the second set of fingerprints. A determination is made whether the encoded version of the content item is synchronized based at least in part on the first curve and the second curve.

Systems and methods for evaluating synchronization between content streams

A distributed video encoding system splits an input video into video segments. The video segments are encoded using multiple video encoding nodes. Prior to the process of splitting the sequence into video segments, the video is analyzed to generate a dependency map. Intelligent segmentation is performed using the dependency map so that each video segment includes all the video frames from which other video frames within that segment have been encoded in the input video. For example, picture headers are inspected to determine the temporal distance of the farthest past and future reference frames used for encoding frames of a video.

Growing community of inventors

Solano Beach, CA, United States of America

Michael Hamilton Coward