Object segmentation method, training method of object segmentation model
CN122336302APending Publication Date: 2026-07-03INSPUR SUZHOU INTELLIGENT TECH CO LTD
View PDF -1 Cites 0 Cited by
Patent Information
- Application Number
- CN202610702225.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-21
- Publication Date
- 2026-07-03
AI Technical Summary
Technical Problem
The existing SAM2 model has high computational redundancy when performing object segmentation and tracking on video streams, making it impossible to achieve real-time processing.
Method used
By introducing preset classification rules, video frames are classified into key frames, regular frames, and frames to be examined. Motion estimation networks and feature enhancement networks are used to accelerate computation and reduce dependence on image encoders.
Benefits of technology
It improves the computational processing capabilities for object segmentation and tracking, reduces computational overhead, and enables real-time processing on video streams.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122336302A_ABST
Abstract
This application provides an object segmentation method and an object segmentation model training method, which can be applied to the field of artificial intelligence technology. The object segmentation method includes: determining whether the current video frame in a video frame sequence is a keyframe, a regular frame, or a frame to be inspected based on a preset classification rule; if the current video frame is a keyframe, extracting features from the current video frame to obtain the current frame image features; if the current video frame is a regular frame or a frame to be inspected, determining the motion change features of the current video frame relative to a reference keyframe associated with the current video frame, and determining the current frame image features based on the motion change features and the reference image features; and performing object segmentation on the current frame image features based on the target mask features representing the target object to obtain the object segmentation result of the current video frame, wherein the object segmentation result indicates the position and shape of the target object.
Need to check novelty before this filing date? Find Prior Art