Abstract: Techniques and solutions are described for executing a video processing task. A video processing task is received that includes one or more operations to be performed on a digital video file and an identifier of the digital video file. The video processing task is divided into subtasks of operations to be performed on fragments of the video, such as fragments having a particular duration. The duration can correspond to a duration used for video streaming Compared with video processing that is performed as a single task, disclosed techniques can provide improved fault tolerance, as only failed tasks need to be reprocessed. Video processing subtasks can be distributed to a plurality of workers, which can further improve fault tolerance, and can increase the computing power available for video processing, including allowing for the use of heterogenous or unreliable workers.

Abstract: A method and system for uploading a media file container from a first device to a second device are described herein, including receiving an instruction to upload the media file container and in response, reading a metadata box of the media file container to locate a track box containing information about audio data, including a size and a location of the audio data, in a media data box of the media file container, identifying the audio data in the media data box using the information from the track box, packaging the identified audio data from the media data box into an audio byte stream separate from the media data box, and uploading the audio byte stream to the second device prior to completing upload of the media file container.

Abstract: Techniques and solutions are described for encoding digital video files, such as for streaming applications. Data associated with the digital video file forms a dataset that can be characterized by a measure of the dataset's center, such as an average, and a spread of the dataset, such as a deviation, with respective to a bitrate over a duration of the digital video file. The measure of center and spread are used to calculate a deviation-adjusted bitrate. A deviation adjusted bitrate can be calculated for the entire digital video file, or for particular subsets of the digital video file, such as for segments of a duration forming units of video streaming Disclosed techniques can provide various advantages, including using a reduced bitrate for video or video portions as compared with an average or static bitrate, for lower-complexity video, or using a higher bitrate for video or video portions for higher-complexity video.

Abstract: A media server uses selective just-in-time (“JIT”) transcoding of media such as video. For example, the media server determines a measure of complexity of a given segment of a given media sequence. The given segment has been encoded at a base bit rate. The media server evaluates a complexity condition for the given segment. As part of evaluating the complexity condition, the media server compares the measure of complexity to a complexity threshold. Based at least in part on whether the complexity condition is satisfied, the media server selects between use of preemptive transcoding and use of JIT transcoding for the given segment at a given target bit rate. In this way, the media server can selectively incur the cost of preemptive transcoding operations for the given segment if JIT transcoding would likely introduce unacceptable delay, and the media server can otherwise use JIT transcoding operations for the given segment.

Abstract: A method and system for uploading a media file container from a first device to a second device are described herein, including receiving an instruction to upload the media file container and in response, reading a metadata box of the media file container to locate a track box containing information about video data in a media data box, identifying sample frames of the video data throughout a duration of the video data in the media data box using information from the track box, packaging the identified sample frames, and uploading the packaged sample frames of the video data prior to completing upload of the media file container.

Abstract: Client-side transcoding of a video segment of a video into a particular version. The video streaming service determines that the video segment of the video is to be transcoded into the particular version. Rather than transcode the video segment itself, the service identifies to a client that is streaming the video that the video segment of the video is to be transcoded into the particular version. The client responds by transcoding the video segment into the particular version, and transmitting the transcoded video segment back to the video streaming service.

Abstract: A method and system for uploading a media file container from a first device to a second device are described herein, including receiving an instruction to upload the media file container and in response, reading a metadata box of the media file container to locate a track box containing information about video data in a media data box, identifying sample frames of the video data throughout a duration of the video data in the media data box using information from the track box, packaging the identified sample frames, and uploading the packaged sample frames of the video data prior to completing upload of the media file container.

Abstract: A method and system for uploading a media file container from a first device to a second device are described herein, including receiving an instruction to upload the media file container and in response, identifying a first portion of the media file container and a last portion of the media file container, each of the first and last portions having a size in bytes and including at least one box of the media file container, and uploading the first portion of the media file container and the last portion of the media file container before the intervening portions of the media file container between the first and last portions.

Abstract: Techniques and solutions are described for executing a video processing task. A video processing task is received that includes one or more operations to be performed on a digital video file and an identifier of the digital video file. The video processing task is divided into subtasks of operations to be performed on fragments of the video, such as fragments having a particular duration. The duration can correspond to a duration used for video streaming Compared with video processing that is performed as a single task, disclosed techniques can provide improved fault tolerance, as only failed tasks need to be reprocessed. Video processing subtasks can be distributed to a plurality of workers, which can further improve fault tolerance, and can increase the computing power available for video processing, including allowing for the use of heterogenous or unreliable workers.

Abstract: A method and system for uploading a media file container from a first device to a second device are described herein, including receiving an instruction to upload the media file container and in response, reading a metadata box of the media file container to locate a track box containing information about audio data, including a size and a location of the audio data, in a media data box of the media file container, identifying the audio data in the media data box using the information from the track box, packaging the identified audio data from the media data box into an audio byte stream separate from the media data box, and uploading the audio byte stream to the second device prior to completing upload of the media file container.

Abstract: Techniques and solutions are described for encoding digital video files, such as for streaming applications. Data associated with the digital video file forms a dataset that can be characterized by a measure of the dataset's center, such as an average, and a spread of the dataset, such as a deviation, with respective to a bitrate over a duration of the digital video file. The measure of center and spread are used to calculate a deviation-adjusted bitrate. A deviation adjusted bitrate can be calculated for the entire digital video file, or for particular subsets of the digital video file, such as for segments of a duration forming units of video streaming Disclosed techniques can provide various advantages, including using a reduced bitrate for video or video portions as compared with an average or static bitrate, for lower-complexity video, or using a higher bitrate for video or video portions for higher-complexity video.

Abstract: The importance of video sections of a video file may be determined from features of the video file. The video file may be decoded to obtain video frames and audio data associated with the video frames. Feature scores for each video frame may be obtained by analyzing features of the video frame or the audio data associated with the video frame based on a local rule, a global rule, or both. The feature scores are further combined to derive a frame importance score for the video frame. Based on the feature scores of the video frames in the video file, the video file may be segmented into video sections of different section importance values.

Abstract: Technologies for a single-pass process for enhancing video quality with temporal smoothing. The process may include providing for user overrides of automatically enhanced video/frame characteristics and providing substantially immediate previews of enhanced video frames to a user. The process may also include detecting a degree of shakiness in a portion of the video, and performing or recommending stabilization based on the detected shakiness.

Abstract: Techniques for identifying prominent subjects in video content based on feature point extraction are described herein. Video files may be processed to detect faces on video frames and extract feature points from the video frames. Some video frames may include detected faces and extracted feature points and other video frames may not include detected faces. Based on the extracted feature points, faces may be inferred on video frames where no face was detected. The inferring may be based on feature points. Additionally, video frames may be arranged into groups and two or more groups may be merged. The merging may be based on some groups including video frames having overlapping feature points. The resulting groups each may identify a subject. A frequency representing a number of video frames where the subject appears may be determined for calculating a prominence score for each of the identified subjects in the video file.

Abstract: This disclosure describes techniques for training models from video data and applying the learned models to identify desirable video data. Video data may be labeled to indicate a semantic category and/or a score indicative of desirability. The video data may be processed to extract low and high level features. A classifier and a scoring model may be trained based on the extracted features. The classifier may estimate a probability that the video data belongs to at least one of the categories in a set of semantic categories. The scoring model may determine a desirability score for the video data. New video data may be processed to extract low and high level features, and feature values may be determined based on the extracted features. The learned classifier and scoring model may be applied to the feature values to determine a desirability score associated with the new video data.

Abstract: Technologies for a single-pass process for enhancing video quality with temporal smoothing. The process may include providing for user overrides of automatically enhanced video/frame characteristics and providing substantially immediate previews of enhanced video frames to a user. The process may also include detecting a degree of shakiness in a portion of the video, and performing or recommending stabilization based on the detected shakiness.

Abstract: Technologies for a single-pass process for enhancing video quality with temporal smoothing. The process may include providing for user overrides of automatically enhanced video/frame characteristics and providing substantially immediate previews of enhanced video frames to a user. The process may also include detecting a degree of shakiness in a portion of the video, and performing or recommending stabilization based on the detected shakiness.

Abstract: This disclosure describes techniques for training models from video data and applying the learned models to identify desirable video data. Video data may be labeled to indicate a semantic category and/or a score indicative of desirability. The video data may be processed to extract low and high level features. A classifier and a scoring model may be trained based on the extracted features. The classifier may estimate a probability that the video data belongs to at least one of the categories in a set of semantic categories. The scoring model may determine a desirability score for the video data. New video data may be processed to extract low and high level features, and feature values may be determined based on the extracted features. The learned classifier and scoring model may be applied to the feature values to determine a desirability score associated with the new video data.

Abstract: Automatic story production is implemented by the utilization of theme scripts with user assets to generate a quality finished product with minimum user input or direction. A user chooses a predesigned theme script to be applied to the user's assets to automatically create a story with a particular look and feel. Metadata and feature information, when available, is automatically gathered from the user assets to personalize the generated story. A user can include additional information and/or alter any aspect of the generated story to further personalize the resultant finished product.

Abstract: Techniques for identifying prominent subjects in video content based on feature point extraction are described herein. Video files may be processed to detect faces on video frames and extract feature points from the video frames. Some video frames may include detected faces and extracted feature points and other video frames may not include detected faces. Based on the extracted feature points, faces may be inferred on video frames where no face was detected. The inferring may be based on feature points. Additionally, video frames may be arranged into groups and two or more groups may be merged. The merging may be based on some groups including video frames having overlapping feature points. The resulting groups each may identify a subject. A frequency representing a number of video frames where the subject appears may be determined for calculating a prominence score for each of the identified subjects in the video file.