Iterative media object compression algorithm optimization using decoupled calibration of perceptual quality algorithms

Patent metadata

Field	Value
Publication number	US-12002261-B2
Application number	US-202218064192-A
Country	US
Kind code	B2
Filing date	Dec 9, 2022
Priority date	May 15, 2020
Publication date	Jun 4, 2024
Grant date	Jun 4, 2024

Abstract

One or more multi-stage optimization iterations are performed with respect to a compression algorithm. A given iteration comprises a first stage in which hyper-parameters of a perceptual quality algorithm are tuned independently of the compression algorithm. A second stage of the iteration comprises tuning hyper-parameters of the compression algorithm using a set of perceptual quality scores generated by the tuned perceptual quality algorithm. The final stage of the iteration comprises performing a compression quality evaluation test on the tuned compression algorithm.

First claim

What is claimed is:

1. A computer-implemented method, comprising:
obtaining, via one or more programmatic interfaces of a network-accessible service of a cloud computing environment, an indication of a factor for classifying a plurality of media objects into importance categories, including a first importance category and a second importance category;
identifying (a) a first set of class-specific tuned parameter values to be used to compress media objects of the first importance category, and (b) a second set of class-specific tuned parameter values to be used to compress media objects of the second importance category, wherein:
the first set of class-specific tuned parameter values have been tuned using example media objects classified as the first importance category, and
the second set of class-specific tuned parameter values have been tuned using example media objects classified as the second importance category;

compressing, prior to a display of a first media object, the first media object using the first set of class-specific tuned parameter values, wherein the first media object is classified, using the factor, as belonging to the first importance category; and
compressing, prior to a display of a second media object, the second media object using the second set of class-specific tuned parameter values, wherein the second media object is classified, using the factor, as belonging to the second importance category.

2. The computer-implemented method as recited in claim 1 , wherein the plurality of media objects comprise images of products available from a web site, and wherein the factor comprises a product category.

3. The computer-implemented method as recited in claim 1 , further comprising:
obtaining, via the one or more programmatic interfaces, an example media object for classifying at least some media objects of the plurality of media objects into a third importance category;
classifying, using at least a search operation for media objects similar to the example media object, a third media object of the plurality of media objects into the third importance category; and
compressing, prior to a display of the third media object, the third media object using a third set of class-specific tuned parameter values, wherein the third set of class-specific tuned parameter values is identified for the third importance category.

4. The computer-implemented method as recited in claim 1 , further comprising:
identifying a default set of tuned parameter values to be used to compress media objects which have not been classified into an importance category of a plurality of importance categories, wherein the plurality of importance categories comprises the first importance category and the second importance category; and
compressing, prior to a display of a third media object, the third media object using the default set of tuned parameter values.

5. The computer-implemented method as recited in claim 1 , further comprising:
receiving, at the network-accessible service, a programmatic request to tune a compression algorithm, wherein the first set of class-specific tuned parameter values is identified in response to the request.

6. The computer-implemented method as recited in claim 1 , wherein the first set of class-specific tuned parameter values comprises a value associated with one or more of: (a) chroma subsampling, (b) block prediction, (c) frequency domain transformation, (d) quantization or (e) run-length encoding.

7. The computer-implemented method as recited in claim 1 , wherein identifying the first set of class-specific tuned parameter values comprises utilizing an evolutionary algorithm.

8. A system, comprising:
one or more computing devices;
wherein the one or more computing devices include instructions that upon execution on or across the one or more computing devices:
obtain, via one or more programmatic interfaces of a network-accessible service of a cloud computing environment, an indication of a factor for classifying a plurality of media objects into importance categories, including a first importance category and a second importance category;
identify (a) a first set of class-specific tuned parameter values to be used to compress media objects of the first importance category, and (b) a second set of class-specific tuned parameter values to be used to compress media objects of the second importance category, wherein:
the first set of class-specific tuned parameter values has been tuned using example media objects classified into the first importance category, and
the second set of class-specific tuned parameter values has been tuned using example media objects classified into the second importance category;

compress, prior to a display of a first media object, the first media object using the first set of class-specific tuned parameter values, wherein the first media object is classified, using the factor, as belonging to the first importance category; and
compress, prior to a display of a second media object, the second media object using the second set of class-specific tuned parameter values, wherein the second media object is classified, using the factor, as belonging to the second importance category.

9. The system as recited in claim 8 , wherein the plurality of media objects comprise images of products available from a web site, and wherein the factor comprises a product category.

10. The system as recited in claim 8 , wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices:
obtain, via the one or more programmatic interfaces, an example media object for classifying at least some media objects of the plurality of media objects into a third importance category;
classify, using at least a search operation for media objects similar to the example media object, a third media object of the plurality of media objects into the third importance category; and
compress, prior to a display of the third media object, the third media object using a third set of class-specific tuned parameter values, wherein the third set of class-specific tuned parameter values is identified for the third importance category.

11. The system as recited in claim 8 , wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices:
identify a default set of tuned parameter values to be used to compress media objects which have not been classified into an importance category of a plurality of importance categories, wherein the plurality of importance categories comprises the first importance category and the second importance category; and
compress, prior to a display of a third media object, the third media object using the default set of tuned parameter values.

12. The system as recited in claim 8 , wherein the one or more computing devices include further instructions that upon execution on or across the one or more computing devices:
receive, at the network-accessible service, a programmatic request to tune a compression algorithm, wherein the first set of class-specific tuned parameter values is identified in response to the request.

13. The system as recited in claim 8 , wherein the first set of class-specific tuned parameter values comprises a value associated with one or more of: (a) chroma subsampling, (b) block prediction, (c) frequency domain transformation, (d) quantization or (e) run-length encoding.

14. The system as recited in claim 8 , wherein to identify the first set of class-specific tuned parameter values, the one or more computing devices inc

Assignees

• Amazon Tech Inc

Inventors

• Staudigl Luitpold
• Wallis Thomas Sydney Austin
• Mueller Mike
• Javed Muhammad Bilal
• Barbachano Pablo

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06V10/82 Primary

  using neural networks · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/251

  of input or preprocessed data · CPC title