Netflix Inc.

Anne Aaron

Jan De Cock

Ioannis Katsavounidis

Zhi Li

Anush Moorthy

Christos G Bampis

David Randall Ronca

Liwei Guo

Andy Schuler

Dae Kim

Kyle Christopher Swanson

Kuyen Tsao

Glenn Van Wallendael

Chi-Hao Wu

Sonia Bhaskar

Yu-Chieh Lin

David Ronca

Zhonghua Ma

Aditya Mavlankar

Todd Goodall

Christos Bampis

In various embodiments, a sequence-based encoding application partitions a set of shot sequences associated with a media title into multiple clusters based on at least one feature that characterizes media content and/or encoded media content associated with the media title. The clusters include at least a first cluster and a second cluster. The sequence-based encoding application encodes a first shot sequence using a first operating point to generate a first encoded shot sequence. The first shot sequence and the first operating point are associated with the first cluster. By contrast, the sequence-based encoding application encodes a second shot sequence using a second operating point to generate a second encoded shot sequence. The second shot sequence and the second operating point are associated with the second cluster. Subsequently, the sequence-based encoding application generates an encoded media sequence based on the first encoded shot sequence and the second encoded shot sequence.

Optimizing encoding operations when generating encoded versions of a media title

In various embodiments, a perceptual quality application determines an absolute quality score for encoded video content viewed on a target viewing device. In operation, the perceptual quality application determines a baseline absolute quality score for the encoded video content viewed on a baseline viewing device. Subsequently, the perceptual quality application determines that a target value for a type of the target viewing device does not match a base value for the type of the baseline viewing device. The perceptual quality application computes an absolute quality score for the encoded video content viewed on the target viewing device based on the baseline absolute quality score and the target value. Because the absolute quality score is independent of the viewing device, the absolute quality score accurately reflects the perceived quality of a wide range of encoded video content when decoded and viewed on a viewing device.

Device-consistent techniques for predicting absolute perceptual video quality

In various embodiments, an encoding ladder application generates encoding ladders for encoding media titles. In operation, the encoding ladder application generates a first convex hull representing encoding tradeoffs between quality and bitrate when encoding a media title at a first resolution; The encoding ladder application generates a second convex hull representing encoding tradeoffs between quality and bitrate when encoding the media title at a second resolution. Based on the first convex hull and the second convex hull, the encoding ladder application generates an overall convex hull. Subsequently, the encoding ladder application generates an encoding ladder for the media title based on at least the overall convex hull and a ladder requirement. Advantageously, the tradeoffs between quality and bitrate represented by the encoding ladder are customized for the media title. Consequently, encoding inefficiencies attributable to conventional fixed-bitrate ladders can be reduced.

Techniques for generating per-title encoding ladders

In various embodiments, a perceptual quality application computes an absolute quality score for encoded video content. In operation, the perceptual quality application selects a model based on the spatial resolution of the video content from which the encoded video content is derived. The model associates a set of objective values for a set of objective quality metrics with an absolute quality score. The perceptual quality application determines a set of target objective values for the objective quality metrics based on the encoded video content. Subsequently, the perceptual quality application computes the absolute quality score for the encoded video content based on the selected model and the set of target objective values. Because the absolute quality score is independent of the quality of the video content, the absolute quality score accurately reflects the perceived quality of a wide range of encoded video content when decoded and viewed.

Source-consistent techniques for predicting absolute perceptual video quality

In various embodiments, an iterative encoding application generates shot encode points based on a first set of encoding points and a first shot sequence associated with a media title. The iterative encoding application performs convex hull operations across the shot encode points to generate a first convex hull. Subsequently, the iterative encoding application generates encoded media sequences based on the first convex hull and a second convex hull that is associated with both a second shot sequence associated with the media title and a second set of encoding points. The iterative encoding application determines a first optimized encoded media and a second optimized encoded media sequence from the encoded media sequences based on, respectively, a first target metric value and a second target metric value for a media metric. Portions of the optimized encoded media sequences are subsequently streamed to endpoint devices during playback of the media title.

Iterative techniques for generating multiple encoded versions of a media title

In various embodiments, an interpolation-based encoding application encodes a first subsequence included in a media title at each encoding point included in a first set of encoding points to generate encoded subsequences. Subsequently, the interpolation-based encoding application performs interpolation operation(s) based on the encoded subsequences to estimate a first media metric value associated with a first encoding point that is not included in the first set of encoding points. The interpolation-based encoding application then generates an encoding recipe based on the encoded subsequences and the first media metric value. The encoding recipe specifies a different encoding point for each subsequence included in the media title. After determining that the encoding recipe specifies the first encoding point for the first subsequence, the interpolation-based encoding application encodes the first subsequence at the first encoding point to generate at least a portion of an encoded version of the media title.

Techniques for efficiently performing subsequence-based encoding for a media title

In various embodiments, a quality trainer trains a model that computes a value for a perceptual video quality metric for encoded video content. During a pre-training phase, the quality trainer partitions baseline values for metrics that describe baseline encoded video content into partitions based on genre. The quality trainer then performs cross-validation operations on the partitions to optimize hyperparameters associated with the model. Subsequently, during a training phase, the quality trainer performs training operations on the model that includes the optimized hyperparameters based on the baseline values for the metrics to generate a trained model. The trained model accurately tracks the video quality for the baseline encoded video content. Further, because the cross-validation operations minimize any potential overfitting, the trained model accurately and consistently predicts perceived video quality for non-baseline encoded video content across a wide range of genres.

Techniques for robustly predicting perceptual video quality

In one embodiment of the present invention, an encode validator identifies and classifies errors introduced during the parallel chunk-based translation of a source to a corresponding aggregate encode. In operation, upon receiving a source for encoding, a frame difference generator creates a frame difference file for the source. A parallel encoder then distributes per-chunk encoding operations across machines and creates an aggregate encode. The encode validator decodes the aggregate encode and creates a corresponding frame difference file. Subsequently, the encode validator performs phase correlation operations between the two frame difference files to detect errors generated by encoding process faults (i.e., dropping a frame, etc.) while suppressing discrepancies inherent in encoding, such as those attributable to low bit-rate encoding. Advantageously, since the encode validator leverages frame difference files, this indirect verification technique enables efficient debugging of parallel encoding processes in which the complete source is unavailable for post-encode analysis.

Techniques for identifying errors introduced during encoding

A bitrate allocation engine allocates bitrates for distributed encoding of source data. Upon receiving a chunk of source data, the bitrate allocation engine generates a curve based on multiple points that each specify a different visual quality level and corresponding encoding bitrate for encoding the chunk. Subsequently, the bitrate allocation engine computes an optimized encoding bitrate based on the generated curve and an optimization factor that is associated with different visual quality levels and corresponding encoding bitrates for multiple chunks of the source data. The bitrate allocation engine then causes the chunk to be encoded at the optimized encoding bitrate. Advantageously, the resulting encoded chunk is optimized with respect to the optimization factor for multiple chunks of the source data.

Growing community of inventors

Menlo Park, CA, United States of America

Anne Aaron