Model training method, video rain removal method, device, equipment and storage medium

By introducing time consistency weights in the video rain removal model training and building a loss function, the problem of poor rain removal effect and fuzzy artifacts when processing large amounts of object motion videos is solved, and a higher quality video rain removal effect is achieved.

CN117392022BActive Publication Date: 2025-05-27SHANGHAI ARTIFICIAL INTELLIGENCE INNOVATION CENT
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311532538.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-11-16
Publication Date
2025-05-27
Estimated Expiration
2043-11-16

AI Technical Summary

Technical Problem

When existing video rain removal technology processes videos containing a large number of objects, there are problems such as poor rain removal effect and blurred artifacts in the generated videos.

Method used

By obtaining the training dataset, including multiple sets of sample videos, each set of videos containing rainy day videos and corresponding rainless videos. For a single pixel point, its first time consistency weight in the rainy video and the second time consistency weight in the rainless video are determined, and the loss function is constructed based on these weights and the video rain-removing model is trained.

Benefits of technology

Effectively improves the video rain removal effect, reduces fuzzy artifacts, and improves the quality of generated videos, especially when dealing with videos with large amounts of objects moving.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117392022B_ABST
    Figure CN117392022B_ABST
Patent Text Reader

Abstract

The present invention discloses a model training method, a video deraining method, a device, an equipment and a storage medium. A training data set is obtained, the training data set includes multiple groups of sample videos, a single group of sample videos includes a rainy day video and a corresponding rainless video; for a single pixel point in the single group of sample videos, the first time consistency weight corresponding to the pixel point in the corresponding rainy day video and the second time consistency weight corresponding to the corresponding rainless video are determined; a video deraining model is trained based on a loss function, wherein the loss function is determined according to the first time consistency weight and the second time consistency weight of each pixel point in each sample video. The existing video deraining technology solves the problems that when processing videos containing a large number of moving objects, the deraining effect is poor and the generated video has blur artifacts, and the video deraining effect is effectively improved, and the quality of the generated video is improved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Real-time video rain removal method based on attention deformation convolution automatic search

    CN112734672A

  • Method and system for removal of rain streak distortion from a video

    US20190050969A1