VIDEO ENCODER AUTOTUNING OF PARAMETERS

1 . A method comprising:

determining an instance of content and a metric to evaluate a quality of an encoding of the instance of content;

extracting a set of features for the instance of content;

performing an optimized search process to evaluate different combinations of encoding parameter values that are used to encode the content to generate instances of encoded content, wherein the instances of encoded content are compared to the metric to determine a next combination of encoding parameter values to use;

selecting an optimal combination of encoding parameter values that is associated with one of the instances of encoded content, wherein the one of the instances of encoded content is selected based a comparison to the metric;

outputting predicted encoding parameter values from a model using model parameters based on an input of the set of features; and

training the model using the optimal combination of encoding parameter values and the predicted encoding parameter values, wherein the model parameters are adjusted in the training.

2 . The method of claim 1 , wherein performing the optimized search process comprises:

iteratively determining combinations of encoding parameter values based on prior sampling of combinations of encoding parameter values and respective comparisons of instances of encoded content to the metric.

3 . The method of claim 2 , wherein prior sampling of combinations of encoding parameter values is used to focus a search for another combination of encoding parameter values.

4 . The method of claim 1 , wherein performing the optimized search process comprises:

building a surrogate model to model an encoder function to encode the instance of content;

using an acquisition function to determine where to sample encoding parameter values next based on the surrogate model;

encoding the instance of content based on the encoding parameter values to generate an instance of encoded content; and

updating the surrogate model based on comparing the instance of encoded content to the metric.

5 . The method of claim 1 , wherein selecting the optimal combination of encoding parameter values comprises:

determining the instance of encoded content that is associated with a highest ranked value of the metric; and

selecting the combination of encoding parameters that were used to generate the instance of encoded content that is associated with the highest ranked value of the metric.

6 . The method of claim 1 , wherein training the model comprises:

inputting the set of features into the model;

outputting predicted encoding parameter values; and

comparing the predicted encoding parameter values to the optimal combination of encoding parameter values to adjust the model parameters of the model based on a difference between the predicted encoding parameter values and the optimal combination of encoding parameter values.

7 . The method of claim 6 , wherein:

the model parameters are trained using a regression training process or classification training process.

8 . The method of claim 1 , further comprising:

using the model to determine predicted encoding parameter values for a new instance of content, wherein the predicted encoding parameter values are used by an encoder to encode the new instance of content.

9 . The method of claim 8 , wherein using the model comprises:

extracting feature values of the new instance of content;

inputting the feature values into the model; and

outputting the predicted encoding parameters for the new instance of content based on the model parameters that were adjusted.

10 . The method of claim 8 , wherein the predicted encoding parameters for the new instance of content are determined without encoding the new instance of content.

11 . The method of claim 8 , wherein the predicted encoding parameters for the new instance of content are used by the encoder to encode the new instance of content a single time for a target bitrate.

12 . The method of claim 8 , further comprising:

encoding the new instance of content using the predicted encoding parameter values.

13 . The method of claim 1 , wherein training the model comprises:

training the model using a condition that sets a bitrate in a plurality of bitrates, wherein the model is trained to output predicted encoding parameter values for bitrates in the plurality of bitrates.

14 . A non-transitory computer-readable storage medium having stored thereon computer executable instructions, which when executed by a computing device, cause the computing device to be operable for:

determining an instance of content and a metric to evaluate a quality of an encoding of the instance of content;

extracting a set of features for the instance of content;

performing an optimized search process to evaluate different combinations of encoding parameter values that are used to encode the content to generate instances of encoded content, wherein the instances of encoded content are compared to the metric to determine a next combination of encoding parameter values to use;

selecting an optimal combination of encoding parameter values that is associated with one of the instances of encoded content, wherein the one of the instances of encoded content is selected based a comparison to the metric;

outputting predicted encoding parameter values from a model using model parameters based on an input of the set of features; and

training the model using the optimal combination of encoding parameter values and the predicted encoding parameter values, wherein the model parameters are adjusted in the training.

15 . A method comprising:

receiving a trained model, wherein the model was trained using an optimized search process that evaluated different combinations of encoding parameter values that are used to encode an instance of content to generate instances of encoded content, wherein the instances of encoded content are compared to a metric to determine a next combination of encoding parameter values to use to determine an optimal combination of encoding parameter values for the instance of content;

extracting feature values of a new instance of content;

inputting the feature values into the trained model to generate predicted encoding parameters; and

encoding the new instance of content using the predicted encoding parameter values.

16 . The method of claim 15 , wherein the predicted encoding parameters are determined without encoding the new instance of content.

17 . The method of claim 15 , wherein the predicted encoding parameters for the new instance of content are used by the encoder to encode the new instance of content a single time for a target bitrate.

18 . The method of claim 15 , further comprising:

inputting a condition that sets a bitrate in a plurality of bitrates, wherein the trained model outputs the predicted encoding parameter values for the bitrate in the plurality of bitrates.

19 . The method of claim 18 , further comprising:

determining an instance of content in the instances of content and the metric to evaluate a quality of an encoding of the instance of content;

extracting a set of features for the instances of content;

performing the optimized search process;

selecting the optimal combination of encoding parameter values that is associated with one of the instances of encoded content, wherein the one of the instances of encoded content is selected based a comparison to the metric;

outputting predicted encoding parameter values from the model using model parameters based on an input of the set of features; and

training the model using the optimal combination of encoding parameter values and the predicted encoding parameter values, wherein the model parameters are adjusted in the training.

20 . The method of claim 19 , wherein performing the optimized search process comprises:

building a surrogate model to model an encoder function to encode the instance of content;

using an acquisition function to determine where to sample encoding parameter values next based on the surrogate model;

encoding the instance of content based on the encoding parameter values to generate an instance of encoded content; and

updating the surrogate model based on comparing the instance of encoded content to the metric.