Saliency-based compression

1. A method, comprising:

obtaining a machine learning model configured to label media files of a given type corresponding to a type of training media samples, wherein the given type of the media files comprises at least one of an image type, an audio type, or a video type;

assigning, by the machine learning model, a label to a training media sample comprising a plurality of elements;

modifying an element of the plurality of elements to generate an updated training media sample;

assigning, by the machine learning model, a label to the updated training media sample;

comparing the label assigned to the training media sample to the label assigned to the updated training media sample;

determining, using the machine learning model, a salience of the element that was modified based on the comparison of the labels;

generating, using the machine learning model, a saliency map based on a plurality of training media samples, the saliency map including a plurality of values corresponding to elements of the saliency map, each element of the plurality of elements corresponding to at least one value, each of the values correspond to the salience of the corresponding element for perception of the media files by a computer determined by the machine learning model, and each of the values indicate an importance of a corresponding element of the saliency map in assigning a label to a media sample;

obtaining a first media sample of the given type, the first media sample having a first sample size;

compressing elements of the first media sample according to the values corresponding to the plurality of elements of the saliency map to generate a second media sample having a second sample size, wherein the second sample size is smaller than the first sample size;

identifying, using the machine learning model, a pattern between the elements of the second media sample and elements of the training media sample; and

assigning, by the machine learning model, a label to the second media sample based on the identified pattern between the elements of the second media sample and the elements of the training media sample.

2. The method of claim 1 , further comprising verifying labeling accuracy of the machine learning model, wherein verifying the labeling accuracy comprises:

labeling, by the machine learning model, the first media sample;

comparing a first label corresponding to the first media sample as labeled by the machine learning model and a second label corresponding to the second media sample as labeled by the machine learning model; and

updating one or more of the values of the saliency map responsive to the first label and the second label being different based on the comparing.

3. The method of claim 1 , further comprising verifying labeling accuracy of the machine learning model, wherein verifying the labeling accuracy comprises:

labeling, by the machine learning model, the first media sample;

comparing a first label corresponding to the first media sample as labeled by the machine learning model and a second label corresponding to the second media sample as labeled by the machine learning model; and

storing the second media sample in a database responsive to the first label and the second label being the same based on the comparing.

4. The method of claim 1 , wherein:

each of the values included in the saliency map describes a degree of compression of a respective element corresponding to a respective encoding parameter.

5. The method of claim 1 , wherein labeling media files by the machine learning model includes at least one of license plate identification, emotion detection, lie detection, background segmentation, facial recognition, or biometrics recognition.

6. One or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause a system to perform operations, the operations comprising:

obtaining a machine learning model configured to label media files of a given type corresponding to a type of training media samples, wherein the given type of the media files comprises at least one of an image type, an audio type, or a video type;

assigning, by the machine learning model, a label to a training media sample comprising a plurality of elements;

modifying an element of the plurality of elements to generate an updated training media sample;

assigning, by the machine learning model, a label to the updated training media sample;

comparing the label assigned to the training media sample to the label assigned to the updated training media sample;

determining, using the machine learning model, a salience of the element that was modified based on the comparison of the labels;

generating, using the machine learning model, a saliency map based on a plurality of training media samples, the saliency map including a plurality of values corresponding to elements of the saliency map, each element of the plurality of elements corresponding to at least one value, each of the values correspond to the salience of the corresponding element for perception of the media files by a computer determined by the machine learning model, and each of the values indicate an importance of a corresponding element of the saliency map in assigning a label to a media sample;

obtaining a first media sample of the given type, the first media sample having a first sample size;

compressing elements of the first media sample according to the values corresponding to the plurality of elements of the saliency map to generate a second media sample having a second sample size, wherein the second sample size is smaller than the first sample size;

identifying, using the machine learning model, a pattern between the elements of the second media sample and elements of the training media sample; and

assigning, by the machine learning model, a label to the second media sample based on the identified pattern between the elements of the second media sample and the elements of the training media sample.

7. The one or more non-transitory computer-readable storage media of claim 6 , further comprising verifying labeling accuracy of the machine learning model, wherein verifying the labeling accuracy comprises:

labeling, by the machine learning model, the first media sample;

comparing a first label corresponding to the first media sample as labeled by the machine learning model and a second label corresponding to the second media sample as labeled by the machine learning model; and

updating one or more of the values of the saliency map responsive to the first label and the second label being different based on the comparing.

8. The one or more non-transitory computer-readable storage media of claim 6 , further comprising verifying labeling accuracy of the machine learning model, wherein verifying the labeling accuracy comprises:

labeling, by the machine learning model, the first media sample;

comparing a first label corresponding to the first media sample as labeled by the machine learning model and a second label corresponding to the second media sample as labeled by the machine learning model; and

storing the second media sample in a database responsive to the first label and the second label being the same based on the comparing.

9. The one or more non-transitory computer-readable storage media of claim 6 , wherein:

each of the values included in the saliency map describes a degree of compression of a respective element corresponding to a respective encoding parameter.

10. The one or more non-transitory computer-readable storage media of claim 6 , wherein labeling media files by the machine learning model includes at least one of license plate identification, emotion detection, lie detection, background segmentation, facial recognition, or biometrics recognition.

11. A system comprising:

one or more processors; and

one or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause the system to perform operations, the operations comprising:

obtaining a machine learning model configured to label media files of a given type corresponding to a type of training media samples, wherein the given type of the media files comprises at least one of an image type, an audio type, or a video type;

assigning, by the machine learning model, a label to a training media sample comprising a plurality of elements;

modifying an element of the plurality of elements to generate an updated training media sample;

assigning, by the machine learning model, a label to the updated training media sample;

comparing the label assigned to the training media sample to the label assigned to the updated training media sample;

determining, using the machine learning model, a salience of the element that was modified based on the comparison of the labels;

generating, using the machine learning model, a saliency map based on a plurality of training media samples, the saliency map including a plurality of values corresponding to elements of the saliency map, each element of the plurality of elements corresponding to at least one value, each of the values correspond to the salience of the corresponding element for perception of the media files by a computer determined by the machine learning model, and each of the values indicate an importance of a corresponding element of the saliency map in assigning a label to a media sample;

obtaining a first media sample of the given type, the first media sample having a first sample size;

compressing elements of the first media sample according to the values corresponding to the plurality of elements of the saliency map to generate a second media sample having a second sample size, wherein the second sample size is smaller than the first sample size; and

assigning, by the machine learning model, a label to the second media sample.

12. The system of claim 11 , further comprising verifying labeling accuracy of the machine learning model, wherein verifying the labeling accuracy comprises:

labeling, by the machine learning model, the first media sample;

comparing a first label corresponding to the first media sample as labeled by the machine learning model and a second label corresponding to the second media sample as labeled by the machine learning model; and

updating one or more of the values of the saliency map responsive to the first label and the second label being different based on the comparing.

13. The system of claim 11 , further comprising verifying labeling accuracy of the machine learning model, wherein verifying the labeling accuracy comprises:

labeling, by the machine learning model, the first media sample;

comparing a first label corresponding to the first media sample as labeled by the machine learning model and a second label corresponding to the second media sample as labeled by the machine learning model; and

storing the second media sample in a database responsive to the first label and the second label being the same based on the comparing.

14. The system of claim 11 , wherein:

each of the values included in the saliency map describes a degree of compression of a respective element corresponding to a respective encoding parameter.

15. The system of claim 11 , wherein labeling media files by the machine learning model includes at least one of license plate identification, emotion detection, lie detection, background segmentation, facial recognition, or biometrics recognition.