Lsi Corporation

Avago Technologies General IP (singapore) Pte. Ltd.

Synamedia Limited

Cheng-Yu Pai

Lowell Leroy Winger

Ossama E A El Badawy

Longji Wang

Stephen Warrington

Techniques for machine learning based coded size estimation in rate control of video encoding are described. An encoder in accordance with various embodiments uses one or more machine learning approaches to facilitate the rate control of video encoding. When training one or more neural network models, the relationship between the coded size and unit characteristics (e.g., picture pixels and/or picture features) is learned from past encoding. The encoder then uses the trained model(s) to estimate the coded size through model inference with improved accuracy. In some embodiments, the trained model(s) are integrated into the encoder for direct model inference. The direct model inference reduces the overhead of referencing application programming interfaces (APIs) provided by a separate machine learning platform, thus making the rate control methods and systems described herein useful in real time video encoding.

Machine learning based coded size estimation in rate control of video encoding

A method of deblocking an input signal is disclosed. The method generally includes the steps of (A) calculating a plurality of transform coefficients corresponding to each of a plurality of blocks in the input signal at baseband, (B) calculating a plurality of quantization parameters based on the transform coefficients, at least one of the quantization parameters corresponding to each respective one of the blocks and (C) generating an output signal by deblocking the input signal based on the quantization parameters.

Baseband signal quantizer estimation

A method of compressed picture reconstruction using a plurality of post-processed reference pictures. The method generally includes the steps of (A) generating a first of a plurality of reconstructed pictures by decoding a first of a plurality of compressed pictures using at least one of a plurality of non-post-processed reference pictures buffered in a reference memory, wherein the compressed pictures are received in an input bitstream, (B) generating a first of a plurality of processed pictures by artifact processing the first reconstructed picture to remove artifacts, (C) buffering in the reference memory both (i) the first reconstructed picture as one of the non-post-processed reference pictures and (ii) the first processed picture as one of the post-processed reference pictures and (D) generating a second of the reconstructed pictures by decoding a second of the compressed pictures using at least one of the post-processed reference pictures buffered in the reference memory.

Compressed non-reference picture reconstruction from post-processed reference pictures

A method for reduced memory and bandwidth motion adaptive video deinterlacing is disclosed. The method generally includes the steps of (A) generating a frame by deinterlacing a current field in a first of a plurality of modes, (B) generating the frame by deinterlacing using both of the current field and an opposite-parity field in a second of the modes and (C) generating the frame be deinterlacing using all of the current field, the opposite-parity field and another field in a third of the modes, wherein the method uses at least two of the modes.

Reduced memory and bandwidth motion adaptive video deinterlacing

A method for motion adaptive video deinterlacing is disclosed. The method generally includes the steps of (A) in a first plurality among a plurality of modes, generating a frame by deinterlacing a current field using a plurality of memory bandwidth configurations and (B) in a second plurality among the modes, generating the frame by deinterlacing the current field using a plurality of linestore configurations.

Growing community of inventors

Brossard, Canada

Cheng-Yu Pai