Method and apparatus for compression of a task output by machine learning

Patent metadata

Field	Value
Publication number	US-12482256-B2
Application number	US-202217845662-A
Country	US
Kind code	B2
Filing date	Jun 21, 2022
Priority date	Jun 22, 2021
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

Disclosed herein are a method and apparatus for distributed image data processing. The method for distributed image data processing includes performing machine learning on an original image to produce a plurality of different task outputs, combining the plurality of task outputs to extract at least one final output, and compressing the final output and transmitting the final output to a server.

First claim

What is claimed is:

1 . A method for generating compressed image data, comprising:
extracting from an original image a plurality of partial regions; generating an extracted image by combining the plurality of partial regions; generating inference data for the plurality of partial regions; and generating compressed image data by encoding the extracted image and the inference data for the plurality of partial regions, wherein the inference data comprises: position data representing a coordinate of a partial region extracted from the original image, and size data representing a size of the partial region extracted from the original image, wherein the position data and the size data are encoded for each of the plurality of partial regions, and wherein a size of the extracted image is different from a size of the original image.

2 . The method of claim 1 , wherein the inference data further comprises information on whether there is a pixel in an extracted partial region.

3 . The method of claim 1 , wherein the method further comprises adjusting a spatial resolution of the extract image, the extract image with an adjusted spatial resolution being encoded to generate the compressed image data.

4 . The method of claim 1 , wherein the plurality of partial regions are extracted by performing a plurality of tasks.

5 . The method of claim 1 , wherein the plurality of partial regions are extracted by a machine learning based on a neural network.

6 . An apparatus for generating compressed image data processing, comprising:
a memory configured to store a control program for generating the compressed image data; and a processor configured to execute the control program stored in the memory, wherein the processor is configured to: extract from an original image a plurality of partial regions; generate an extracted image by combining the plurality of partial regions; generate inference data for the plurality of partial regions; and generate the compressed image data by encoding the extracted image and the inference data for the plurality of partial regions, wherein the inference data comprises: position data representing a coordinate of a partial region extracted from the original image, and size data representing a size of the partial region extracted from the original image, wherein the position data and the size data are encoded for each of the plurality of partial regions, and wherein a size of the extracted image is different from a size of the original image.

7 . A method for decompressing compressed image data, comprising:
decoding an extracted image from the compressed image data, the extracted image comprising a plurality of partial regions; obtaining inference data for the plurality of partial regions in the extracted image; and generating an output image from the extracted image, wherein the inference data comprises: position data representing a coordinate of a partial region included in the extracted image, and size data representing a size of the partial region included in the extracted image, wherein the position data and the size data are obtained for each of the plurality of partial regions, and wherein a size of the extracted image is different from a size of the output image.

Assignees

• Electronics & Telecommunications Res Inst
• Myongji Univ Industry And Academia Cooperation Foundation

Inventors

• Lee Jin-Young
• Lee Hee-Kyung
• Kim Sang-Kyun

Classifications

• G06V10/82 Primary

  using neural networks · CPC title

• G06V10/22

  by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition · CPC title

• G06V10/56

  relating to colour · CPC title

• G06V10/94

  Hardware or software architectures specially adapted for image or video understanding · CPC title

• G06V10/80

  Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level (multimodal speaker identification or verification G10L17/10) · CPC title