US 12,219,180 B2
Actively-learned context modeling for image compression
Gang Wu, San Jose, CA (US); Yang Li, Chapel Hill, NC (US); Stefano Petrangeli, Mountain View, CA (US); Viswanathan Swaminathan, Saratoga, CA (US); Haoliang Wang, San Carlos, CA (US); Ryan A. Rossi, San Jose, CA (US); and Zhao Song, Seattle, WA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on May 20, 2022, as Appl. No. 17/749,846.
Prior Publication US 2023/0379507 A1, Nov. 23, 2023
Int. Cl. G06K 9/00 (2022.01); G06N 20/00 (2019.01); H04N 19/182 (2014.01); H04N 19/184 (2014.01); H04N 19/50 (2014.01); H04N 19/91 (2014.01); H04N 19/96 (2014.01)
CPC H04N 19/96 (2014.11) [G06N 20/00 (2019.01); H04N 19/182 (2014.11); H04N 19/184 (2014.11); H04N 19/50 (2014.11); H04N 19/91 (2014.11)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for actively-learned context modeling, the method comprising:
selecting a subset of data from a training dataset corresponding with an image to be compressed, the subset of data corresponding with a subset of data of pixels of the image;
generating a context model using the selected subset of data, the context model in the form of a decision tree having a set of leaf nodes;
determining entropy values corresponding with each leaf node of the set of leaf nodes, each entropy value indicating an extent of diversity of context associated with the corresponding leaf node;
selecting additional data from the training dataset based on the entropy values corresponding with the leaf nodes, the additional data being used to update the subset of data; and
using the updated subset of data to generate an updated context model for use in performing compression of the image.
 
12. A computer-implemented method for actively-learned context modeling, the method comprising:
selecting a subset of data from a training dataset corresponding with an image to be compressed, the subset of data corresponding with a subset of pixels of the image;
generating a context model using the subset of data, the context model in the form of a decision tree having a set of leaf nodes;
iteratively updating the subset of data by adding new data from the training dataset to the subset of data, wherein the new data to add from the training dataset is selectively identified in each iteration by:
determining a likelihood of a prediction value for a portion of the training dataset absent in the subset of data;
selecting the new data from the portion of the training dataset absent in the subset of data, the new data having a highest likelihood of the prediction value;
updating the subset of data with the additional data; and
using the updated subset of data to generate an updated context model for use in performing compression of the image.
 
20. A system comprising:
a memory device; and
a processing device, operatively coupled to the memory device, to perform operations comprising:
selecting a subset of data from a training dataset corresponding with an image to be compressed, the subset of data corresponding with a subset of data of pixels of the image;
iteratively updating the subset of data by adding new data from the training dataset to the subset of data, wherein the new data to add from the training dataset is selectively identified in each iteration by:
generating an updated context model using the updated subset of data, the updated context model in the form of a decision tree having a set of leaf nodes;
determining entropy values corresponding with each leaf node of the set of leaf nodes, each entropy value indicating an extent of diversity of context associated with the corresponding leaf node; and
selecting the new data from the training dataset based on the entropy values corresponding with the leaf nodes, the new data being used to update the subset of data; and
based on the subset of data attaining a threshold subset of data size, using the corresponding updated subset of data to generate a final context model for use in performing compression of the image.