Image super-resolution model processing method and device, electronic equipment and chip

Through reinforcement learning model, it provides cropping strategies for image super-resolution models, solves the problem of high hardware requirements for large-scale image super-resolution models, realizes the lightweight processing of the model, reduces memory and computing costs, and is suitable for resource-limited resources equipment.

CN119048345APending Publication Date: 2024-11-29BEIJING XIAOMI MOBILE SOFTWARE CO LTD
-1 Cites 0 Cited by

Patent Information

Application Number
CN202310587480.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-23
Publication Date
2024-11-29

Smart Images

  • Figure CN119048345A_ABST
    Figure CN119048345A_ABST
Patent Text Reader

Abstract

The invention discloses a processing method and device of an image super-resolution model, electronic equipment and a chip, and relates to the technical field of image data processing, and the method comprises the steps: firstly obtaining an image super-resolution model to be cut, and a reinforcement learning model used for providing a cutting strategy for the image super-resolution model; determining feature data of each network block in the image super-resolution model; and then, according to the feature data of each network block, sequentially performing iteration processing on each network block by using the image super-resolution model to obtain cutting strategy information of each network block, and based on the cutting strategy information of each network block, performing lightweight processing on the image super-resolution model. According to the technical scheme, lightweight processing can be effectively carried out on a complex large-scale image super-resolution model, a lightweight model is obtained, the requirements for hardware such as a CPU and a GPU of computer equipment are reduced, the occupied memory size can be reduced, and the calculation cost can be effectively saved.
Need to check novelty before this filing date? Find Prior Art