Transform-based grabbing pose detection method
The invention discloses a grabbing pose detection method based on Transform, and belongs to the technical field of robot identification, and the method comprises the steps: sending an obtained input image into an encoder for feature extraction, and outputting a feature map, the encoder comprising a...
Saved in:
Main Authors | , , , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
17.10.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention discloses a grabbing pose detection method based on Transform, and belongs to the technical field of robot identification, and the method comprises the steps: sending an obtained input image into an encoder for feature extraction, and outputting a feature map, the encoder comprising a plurality of Swin conversion modules; the first decoder performs up-sampling and feature fusion on the feature pattern; the second decoder decodes the feature map and outputs a captured quality score map, a width map and an angle map; a grabbing pose detection model based on a Transform is constructed; constructing a loss function, taking grabbing posture prediction as a regression problem, training and optimizing a grabbing posture detection model by minimizing the distance between a predicted grabbing image and a real grabbing image until the training of the grabbing posture detection model is converged, and finally obtaining a Transform-based grabbing posture detection model; and carrying out grabbing position a |
---|---|
Bibliography: | Application Number: CN202311078801 |