Zero sample target recognition method based on Transform-VAE
A target recognition and sample technology, applied in the field of deep learning image recognition, can solve problems such as high labor and time costs, difficulty in obtaining a large amount of image data, and inability to achieve image recognition, so as to promote the development of the field, improve task performance, and reduce acquisition. The effect of labeling costs
Pending Publication Date: 2021-11-23
ZHEJIANG LAB
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
In practical applications, obtaining a large amount of labeled data requires high manpower and time costs, and for uncommon categories, it is difficult to obtain a large amount of image data, so that it is impossible to realize its image recognition
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0017] In order to make the purpose, technical solution and technical effect of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings.
[0018] In the zero-shot target recognition task, the categories of images, image category labels, and category semantic word vectors can be obtained for known classes, and the categories of unknown classes can only be obtained for category semantic word vectors, and there is no intersection between known classes and unknown classes.
[0019] Such as figure 1 Shown, a kind of Transformer-VAE-based zero sample target recognition method of the present invention comprises the following steps:
[0020] Step 1, the generation of image visual features of known classes and the encoding and decoding of visual features: Input the images of known classes into the pre-trained network on ImageNet to obtain the corresponding visual features of the image, and then encode ...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention relates to the field of deep learning image recognition, in particular to a zero-sample target recognition method based on Transform-VAE, which combines an encoder and a decoder of Transform and an encoder and a decoder of VAE to realize encoding and decoding of a visual feature domain and encoding and decoding of a semantic feature domain, and through cross-domain alignment loss constraint, the hidden variables of the visual feature domain and the semantic feature domain are limited in the same space, and the hidden variables obtained by encoding the known class and the unknown class are classified to realize zero sample target recognition. Zero sample target recognition can be well generalized to an unknown class after being trained on a known class, so that the sample collection and labeling cost is greatly reduced, the task performance of image recognition in a scene without samples and with few samples is greatly improved, the field development of zero sample target recognition is promoted, and the research and application of zero sample target identification in scientific research and industry are accelerated.
Description
technical field [0001] The invention relates to the field of deep learning image recognition, in particular to a zero-sample target recognition method based on Transformer-VAE. Background technique [0002] With the development of computer vision and image technology, deep learning has been widely used in various fields such as image classification, target detection, and image segmentation due to its high performance. A large amount of labeled data is the prerequisite and necessary condition for deep learning algorithms to achieve high performance. In practical applications, obtaining a large amount of labeled data requires high manpower and time costs, and for uncommon categories, it is difficult to obtain a large amount of image data, so that the recognition of its images cannot be realized. Therefore, few-shot learning and zero-shot learning have attracted extensive attention and research. [0003] Zero-sample object recognition technology, that is, after training on th...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/46G06K9/62
CPCG06F18/241
Inventor 刘亚洁宋明黎
Owner ZHEJIANG LAB