Gesture segmentation network device and method based on multi-branch cascade Transformer
The invention provides a gesture segmentation network device and method based on a multi-branch cascade Transform, and the device comprises a degree convolutional neural network DCNN which carries out the feature extraction of an original gesture image, and obtains an intermediate feature map. The i...
Saved in:
Main Authors | , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
25.11.2022
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention provides a gesture segmentation network device and method based on a multi-branch cascade Transform, and the device comprises a degree convolutional neural network DCNN which carries out the feature extraction of an original gesture image, and obtains an intermediate feature map. The invention discloses a multi-branch cascade Transformation module (MBCT), which comprises a plurality of Transformation branches which are cascaded together, and each Transformation branch comprises a patch partition layer, a linear embedding layer and a multi-window self-attention block (MWSA) which are connected together in series. And the decoder is used for recovering to the same size as the original gesture image. The hand edge part of the segmentation result is smoother, the capability of removing complex backgrounds is higher, and the robustness and the effectiveness are higher. The method has high accuracy and high robustness under the conditions of uneven illumination, complex background noise, changeable ge |
---|---|
Bibliography: | Application Number: CN202210830586 |