Apple Inc.

Amazon Technologies, Inc.

Ke Zhang

Hsi-Jung Wu

Xiaosong Zhou

Chris Yoochang Chung

Jiefu Zhai

Shujie Liu

Yeping Su

Albert E Keinath

Yunfei Zheng

Hao Pan

Douglas Scott Price

James O Normile

Mark Kamichoff

Dazhong Zhang

Ravi Manghirmalani

Colin John Whittaker

Vinay Sharma

Emilie Kim

Tahsin Erdogan

Bartlomiej W Rymkowski

James Paul Rivers

Dongling Duan

Leonid Livak

Ethan J Tira-Thompson

Shuang Gao

Robert A Bailey

Yuan Yuan

Xiaoxia Sun

Chinchu Merine Joseph

Ben Englert

Shrawan Chittoor Surender

Shriram Srinivasan

Abhishek Vinayak Naik

A host-to-ToR switch advertisement protocol can be used at a data center network edge to receive IP advertisements from hosts and convey ToR switch availability status (e.g., in service, intentionally out of service, or unintentionally out of service) back to hosts and vice versa. The host server computers can have the ability to advertise an arbitrary number of IPv4 or IPv6 prefixes to the network through the ToR switches. A single host server computer can leverage multiple instances across multiple ports to the same switch or different switches to implement load balancing (e.g., equal cost multipath (ECMP)) and fault tolerance. The host server computers benefit from an increased availability by receiving network traffic via multiple ToR switches, a single ToR switch with multiple links, or a combination of both.

Network rack of host server computers with increased network availability

Video object and keypoint location detection techniques are presented. The system includes a detection system for generation locations of an object's keypoints along with probabilities associated with the locations, and a stability system for stabilizing keypoint locations of the detected objects. In some aspects, the generated probabilities are two-dimensional array correspond locations within input images, and stability system fits the generated probabilities to a two-dimensional probability distribution function.

Object and keypoint detection system with low spatial jitter, low latency and low power usage

Techniques disclosed for managing video captured by an imaging device. Methods disclosed capture a video in response to a capture command received at the imaging device. Following a video capture, techniques for classifying the captured video based on feature(s) extracted therefrom, for marking the captured video based on the classification, and for generating a media item from the captured video according to the marking are disclosed. Accordingly, the captured video may be classified as representing a static event, and, as a result, a media item of a still image may be generated. Otherwise, the captured video may be classified as representing a dynamic event, and, as a result, a media item of a video may be generated.

Automated media editing operations in consumer devices

An extension is provided to a Link Aggregation Control Protocol (LACP) that can use the LACP protocol transport and provides Link Aggregation Group (LAG) management while also functioning as a full liveness detection protocol. Bi-directional link detection is supported and timers are configurable to any number. The extension can be backwards compatible with standard LACP and can use a subtype that is specified as unused. The extension can start up using standard LACP packet rates and include additional information in the unused subtype. If a LACP speaker does not support the extension, then the protocol conforms to standard LACP. A state machine of the extension is used if it detects conforming information from a peer speaker. The state machine can allow faster detection should a link error occur.

Network link liveness detection

Systems and methods are provided for processing high quality video data, such as data having a higher than standard bit depth, a high dynamic range, or a wide or custom color gamut, to be compatible with conventional encoders and decoders without significant loss of quality. High quality data is encoded into a plurality of layers with a base layer having the standard quality data and one or more higher quality layers. Decoding systems and methods may map the base layer to the dynamic range or color gamut of the enhancement layer, combine the layers, and map the combined layers to a dynamic range or color gamut appropriate for the target display. Each of the standard quality and the high quality data may be encoded as a plurality of tiers of increasing quality and reference lower level tiers as sources of prediction during predictive coding.

Techniques for HDR/WCR video coding

Systems and processes for improved video editing, summarization and navigation based on generation and analysis of metadata are described. The metadata may be content-based (e.g., differences between neighboring frames, exposure data, key frame identification data, motion data, or face detection data) or non-content-based (e.g., exposure, focus, location, time) and used to prioritize and/or classify portions of video. The metadata may be generated at the time of image capture or during post-processing. Prioritization information, such as a score for various portions of the image data may be based on the metadata and/or image data. Classification information such as the type or quality of a scene may be determined based on the metadata and/or image data. The classification and prioritization information may be metadata and may be used to automatically remove undesirable portions of the video, generate suggestions during editing or automatically generate summary video.

Video analysis techniques for improved editing, navigation, and summarization

Techniques are disclosed for estimating quality of images in an automated fashion. According to these techniques, a source image may be downsampled to generate at least two downsampled images at different levels of downsampling. Blurriness of the images may be estimated starting with a most-heavily downsampled image. Blocks of a given image may be evaluated for blurriness and, when a block of a given image is estimated to be blurry, the block of the image and co-located blocks of higher resolution image(s) may be designated as blurry. Thereafter, a blurriness score may be calculated for the source image from the number of blocks of the source image designated as blurry.

Hierarchical sharpness evaluation

A system for processing media on a resource restricted device, the system including a memory to store data representing media assets and associated descriptors, and program instructions representing an application and a media processing system, and a processor to execute the program instructions, wherein the program instructions represent the media processing system, in response to a call from an application defining a plurality of services to be performed on an asset, determine a tiered schedule of processing operations to be performed upon the asset based on a processing budget associated therewith, and iteratively execute the processing operations on a tier-by-tier basis, unless interrupted.

Media analysis and processing framework on a resource restricted device

Techniques for cropping images containing an occlusion are presented. A method for image editing is presented comprising, when an occlusion is detected in an original digital image, determining an area occupied by the occlusion, assigning importance scores to different content elements of the original digital image, defining a cropping window around an area of the original digital image that does not include the area occupied by the occlusion and that is based on the importance scores, and cropping the original digital image to the cropping window.

Image/video editor with automatic occlusion detection and cropping

A method for processing media assets includes, given a first media asset, deriving characteristics from the first media asset, searching for other media assets having characteristics that correlate to the characteristics of the first media asset, when a match is found, deriving content corrections for the first media asset or a matching media asset from the other of the first media asset or the matching media asset, and correcting content of the first media asset or the matching media asset based on the content corrections.

Cross-asset media analysis and processing

Methods for organizing media data by automatically segmenting media data into hierarchical layers of scenes are described. The media data may include metadata and content having still image, video or audio data. The metadata may be content-based (e.g., differences between neighboring frames, exposure data, key frame identification data, motion data, or face detection data) or non-content-based (e.g., exposure, focus, location, time) and used to prioritize and/or classify portions of video. The metadata may be generated at the time of image capture or during post-processing. Prioritization information, such as a score for various portions of the image data may be based on the metadata and/or image data. Classification information such as the type or quality of a scene may be determined based on the metadata and/or image data. The classification and prioritization information may be metadata and may be used to organize the media data.

Hierarchical segmentation and quality measurement for video editing

A scalable coding system codes video as a base layer representation and an enhancement layer representation. A base layer coder may code an LDR representation of a source video. A predictor may predict an HDR representation of the source video from the coded base layer data. A comparator may generate prediction residuals which represent a difference between an HDR representation of the source video and the predicted HDR representation of the source video. A quantizer may quantize the residuals down to an LDR representation. An enhancement layer coder may code the LDR residuals. In other scalable coding systems, the enhancement layer coder may code LDR-converted HDR video directly.

Techniques in backwards compatible multi-layer compression of HDR video

Video coding systems and methods are provided that protect against various artifacts (e.g., banding artifacts) in decoded frame content. The embodiments to protect against introducing artifacts include pre-processing the input video frame, controlling the encoding parameters and/or post-processing the decoded video frame. A method for encoding a frame may include calculating a dark-scene score for the frame based on measured statistics of the frame. The dark-scene score may be compared to a threshold to determine if the banding artifacts may be present in the frame. Based on the comparison, filtering parameters for the frame may be altered to reduce the amount of noise filtered from the frame. The frame may be filtered based on the altered filtering parameters and encoded.

Video processors for preserving detail in low-light scenes

An encoder may include a luma transform, a transformer, and a chroma transform. The luma transform may determine a linear luminance value based upon a plurality of primary color values of a pixel. The transformer may generate a transformed luminance value based upon the linear luminance value and a plurality of transformed color values based upon corresponding more than one of the primary color values of the pixel. The chroma transform may determine a plurality of chroma values based upon corresponding plurality of transformed color values and the transformed luminance value of the pixel.

YUV signal generation for high dynamic range video content

A video coding server may code a common video sequence into a plurality of coded data streams, each coded data stream representing the video sequence coded using coding parameters tailored for a respective transmission bit rate. The coding may cause a set of transmission units from among the coded data streams to include coded video data from a common point of the video sequence, and a first coded frame of each transmission unit of the set to be a synchronization frame. A manifest file may be built representing an index of transmission units of the respective coded data streams. The coded data streams and manifest file may be stored by the server for delivery to a client device. During download and decode, the chunks may be decoded efficiently even when switching among streams because the first frame in each chunk is a synchronization frame.

Growing community of inventors

San Jose, CA, United States of America

Ke Zhang