DEEP LEARNING-BASED QUALITY CONTROL OF VIDEO COMPRESSION

1 . A computer-implemented method for real-time quality control of video compression for streaming media, comprising:

partitioning video data from the streaming media into video chunks;

encoding the video chunks with respective quantization parameters (QP);

predicting an optimized QP from the video chunks and their respective QP that conforms to a peak signal-to-noise ratio (PSNR) threshold while minimizing an encoded video bitrate of the compressed video by utilizing a trained video quality control unit (VQCU); and

updating the video quality of a streaming media with the optimized QP.

2 . The computer-implemented method of claim 1 , further comprising providing the streaming media having the updated video quality with the optimized QP to a streaming media device through a network.

3 . The computer-implemented method of claim 1 , wherein predicting the optimized QP further comprises training the VQCU using tuples of the video chunks, corresponding quantization parameters, and a minimum PSNR threshold.

4 . The computer-implemented method of claim 1 , wherein predicting the optimized QP further comprises understanding scene complexity and video dynamics by using a video dynamics neural network.

5 . The computer-implemented method of claim 4 , wherein predicting the optimized QP further comprises generating feature vectors representing scene complexity and video dynamics using convolutional layers of the video dynamics neural network.

6 . The computer-implemented method of claim 5 , wherein predicting the optimized QP further comprises normalizing the feature vectors using a deep neural network prediction head.

7 . The computer-implemented method of claim 6 , wherein predicting the optimized QP further comprises generating a tensor representing a minimum PSNR for each video sample for a conditional group normalization block of the deep neural network prediction head.

8 . A system for real-time quality control of video compression for streaming media, comprising:

a memory device;

one or more processor devices operatively coupled with the memory device to:

partition video data from the streaming media into video chunks;

encode the video chunks with respective quantization parameters (QP);

predict an optimized QP from the video chunks and their respective QP that conforms to a peak signal-to-noise ratio (PSNR) threshold while minimizing an encoded video bitrate of the compressed video by utilizing a trained video quality control unit (VQCU); and

update the video quality of a streaming media with the optimized QP.

9 . The system of claim 8 , further comprising providing the streaming media having the updated video quality with the optimized QP to a streaming media device through a network.

10 . The system of claim 8 , wherein to predict the optimized QP further comprises training the VQCU using tuples of the video chunks, corresponding quantization parameters, and a minimum PSNR threshold.

11 . The system of claim 8 , wherein to predict the optimized QP further comprises understanding scene complexity and video dynamics by using a video dynamics neural network.

12 . The system of claim 11 , wherein to predict the optimized QP further comprises to generate feature vectors representing scene complexity and video dynamics using convolutional layers of the video dynamics neural network.

13 . The system of claim 12 , wherein to predict the optimized QP further comprises to normalize the feature vectors using a deep neural network prediction head.

14 . The system of claim 13 , wherein to predict the optimized QP further comprises to generate tensors representing a minimum PSNR for each video sample for a conditional group normalization block of the deep neural network prediction head.

15 . A non-transitory computer program product comprising a computer-readable storage medium including program code for deep learning-based quality control for video compression, wherein the program code when executed on a computer causes the computer to:

partition video data from streaming media into video chunks;

encode the video chunks with respective quantization parameters (QP);

predict an optimized QP from the video chunks and their respective QP that conforms to a peak signal-to-noise ratio (PSNR) threshold while minimizing an encoded video bitrate of the compressed video by utilizing a trained video quality control unit (VQCU); and

update the video quality of a streaming media with the optimized QP.

16 . The non-transitory computer program product of claim 15 , further comprising providing the streaming media having the updated video quality with the optimized QP to a streaming media device through a network.

17 . The non-transitory computer program product of claim 15 , wherein to predict the optimized QP further comprises training the VQCU using tuples of the video chunks, corresponding quantization parameters, and a minimum PSNR threshold.

18 . The non-transitory computer program product of claim 15 , wherein to predict the optimized QP further comprises understanding scene complexity and video dynamics by using a video dynamics neural network.

19 . The non-transitory computer program product of claim 18 , wherein to predict the optimized QP further comprises to generate feature vectors representing scene complexity and video dynamics using convolutional layers of the video dynamics neural network.

20 . The non-transitory computer program product of claim 19 , wherein to predict the optimized QP further comprises to normalize the feature vectors using a deep neural network prediction head having a conditional group normalization block that utilizes tensors representing a minimum PSNR for each video sample.