Abstract: Techniques for predicting video encoding complexity are described herein. A method for predicting video encoding complexity includes performing video complexity feature extraction on a video segment to extract low-complexity frame-based features, predicting video encoding complexity for the video segment using the low-complexity frame-based features, and outputting a predicted encoding bitrate and a predicted encoding time. An embodiment may include implementing a hybrid model using a CNN, wherein a latent vector from a frame of the video segment is extracted and also may be used to predict video encoding complexity. The predicted encoding bitrates and encoding times may be provided to encoding infrastructure for use in optimizing a schedule of encodings.

Abstract: Techniques for efficient two-pass encoding for live streaming are described herein. A method for efficient two-pass encoding may include extracting low-complexity features of a video segment, predicting an optimized constant rate factor (CRF) for the video segment using the low-complexity features, and encoding the video segment with the optimized CRF at a target bitrate. A system for efficient two-pass encoding may include a feature extraction module configured to extract low-complexity features from a video segment, a neural network configured to predict an optimized CRF as a function of the low-complexity features and a target bitrate, and an encoder configured to encode the video segment using the optimized CRF at the target bitrate.

Abstract: Techniques for content-adaptive encoder preset prediction for adaptive live streaming are described herein. A method for content-adaptive encoder preset prediction for adaptive live streaming includes performing video complexity feature extraction on a video segment to extract complexity features such as an average texture energy, an average temporal energy, and an average lumiscence. These inputs may be provided to an encoding time prediction model, along with a bitrate ladder, a resolution set, a target video encoding speed, and a number of CPU threads for the video segment, to predict an encoding time, and an optimized encoding preset may be selected for the video segment by a preset selection function using the predicted encoding time. The video segment may be encoded according to the optimized encoding preset.

Abstract: According to embodiments of the disclosure, fast multi-rate encoding may be performed using machine learning by encoding a lowest quality representation to determine encoding parameters, processing raw data of the video using a neural network to obtain an intermediate output comprising encoding features, augmenting the intermediate output with additional encoding features to form a final tensor, and processing the final tensor with another neural network to obtain a classification output comprising a split or not split decision for an image data block. The classification output may be used to encode a highest quality representation, and then other representations of the video.

Abstract: The technology described herein relates to variable framerate encoding. A method for variable framerate encoding includes receiving shots, as segmented from a video input, extracting features for each of the shots, the features including at least a spatial energy feature and an average temporal energy, predicting a frame dropping factor for each of the shots based on the spatial energy feature and the average temporal energy, predicting an optimized framerate for each of the shots based on the frame dropping factor, downscaling and encoding each of the shots using the optimized framerate. The encoded shots may then be decoded and upscaled back to their original framerates.

Abstract: The technology described herein relates to online per-title encoding. A method for online per-title encoding includes receiving a video input, generating segments of the video input, extracting a spatial feature and a temporal feature, predicting bitrate-resolution pairs based on the spatial feature and the temporal feature, using a discrete cosine transform (DCT)-based energy function, and per-title encoding segments of the video input for the predicted bitrate-resolution pairs. A system for online per-title encoding may include memory for storing a set of bitrates, a set of resolutions, and a machine learning module configured to predict bitrate resolution pairs based on low-complexity spatial and temporal features.

Abstract: Techniques for implementing perceptually aware per-title encoding may include receiving an input video, a set of resolutions, a maximum target bitrate and a minimum target bitrate, extracting content aware features for each segment of the input video, predicting a perceptually aware bitrate-resolution pair for each segment using a model configured to optimize for a quality metric using constants trained for each of the set of resolutions, generating a target encoding set including a set of perceptually aware bitrate-resolution pairs, and encoding the target encoding set. The content aware features may include a spatial energy feature and an average temporal energy. According to these methods only a subset of bitrates and resolutions, less than a full set of bitrates and resolutions, are encoded to provide high quality video content for streaming.

Abstract: The technology described herein relates to variable framerate encoding. A method for variable framerate encoding includes receiving shots, as segmented from a video input, extracting features for each of the shots, the features including at least a spatial energy feature and an average temporal energy, predicting a frame dropping factor for each of the shots based on the spatial energy feature and the average temporal energy, predicting an optimized framerate for each of the shots based on the frame dropping factor, downscaling and encoding each of the shots using the optimized framerate. The encoded shots may then be decoded and upscaled back to their original framerates.

Abstract: The technology described herein relates to online per-title encoding. A method for online per-title encoding includes receiving a video input, generating segments of the video input, extracting a spatial feature and a temporal feature, predicting bitrate-resolution pairs based on the spatial feature and the temporal feature, using a discrete cosine transform (DCT)-based energy function, and per-title encoding segments of the video input for the predicted bitrate-resolution pairs. A system for online per-title encoding may include memory for storing a set of bitrates, a set of resolutions, and a machine learning module configured to predict bitrate resolution pairs based on low-complexity spatial and temporal features.

Abstract: A scalable per-title encoding technique may include detecting scene cuts in an input video received by an encoding network or system, generating segments of the input video, performing per-title encoding of a segment of the input video, training a deep neural network (DNN) for each representation of the segment, thereby generating a trained DNN, compressing the trained DNN, thereby generating a compressed trained DNN, and generating an enhanced bitrate ladder including metadata comprising the compressed trained DNN. In some embodiments, the method also may include generating a base layer bitrate ladder for CPU devices, and providing the enhanced bitrate ladder for GPU-available devices.

Abstract: Techniques relating to per-title encoding using spatial and temporal resolution downscaling is disclosed. A method for per-title encoding includes receiving a video input comprised of video segments, spatially downscaling the video input, temporally downscaling the video input, encoding the video input to generate an encoded video, then temporally and spatially upscaling the encoded video. Spatially downscaling may include reducing a resolution of the video input, and temporally downscaling may include reducing a framerate of the video input. Objective metrics for the upscaled encoded video show improved quality over conventional methods.

Abstract: According to embodiments of the disclosure, information of higher and lower quality encoded video segments is used to limit Rate-Distortion Optimization (RDO) for each Coding Unit Tree (CTU). A method first encodes the highest bit-rate segment and consequently uses it to encode the lowest bit-rate video segment. Block structure and selected reference frame of both highest and lowest bit-rate video segments are used to predict and shorten RDO process for each CTU in middle bit-rates. The method delays just one frame using parallel processing. This approach provides time-complexity reduction compared to the reference software for middle bit-rates while degradation is negligible.

Abstract: Disclosed are systems and methods for lightweight transcoding of video. A distributed computing system for lightweight transcoding includes an origin server and an edge node, the origin server having a memory and a processor and configured to receive an input video comprising a bitstream, encode the bitstream into a set of representations corresponding to a full bitrate ladder, generate encoding metadata for the set of representations, and provide a representation and encoding metadata for the set of representations to an edge node, the edge node having a memory and a processor and configured to transcode the bitstream, or segments thereof, into the set of representations, and to serve one or more of the representations to a client.

Abstract: According to embodiments of the disclosure, fast multi-rate encoding may be performed using machine learning by encoding a lowest quality representation to determine encoding parameters, processing raw data of the video using a neural network to obtain an intermediate output comprising encoding features, augmenting the intermediate output with additional encoding features to form a final tensor, and processing the final tensor with another neural network to obtain a classification output comprising a split or not split decision for an image data block. The classification output may be used to encode a highest quality representation, and then other representations of the video.

Abstract: According to embodiments of the disclosure, information of higher and lower quality encoded video segments is used to limit Rate-Distortion Optimization (RDO) for each Coding Unit Tree (CTU). A method first encodes the highest bit-rate segment and consequently uses it to encode the lowest bit-rate video segment. Block structure and selected reference frame of both highest and lowest bit-rate video segments are used to predict and shorten RDO process for each CTU in middle bit-rates. The method delays just one frame using parallel processing. This approach provides time-complexity reduction compared to the reference software for middle bit-rates while degradation is negligible.