Unlock instant, AI-driven research and patent intelligence for your innovation.

Lightweight zero sample learning algorithm framework based on attribute knowledge

A sample learning and lightweight technology, applied in the field of image recognition, can solve the problem of increasing model training parameters, and achieve the effect of alleviating the problem and alleviating domain drift

Pending Publication Date: 2022-07-29
TIANJIN UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] At present, the algorithm of attribute-based zero-shot learning has achieved high recognition accuracy, but the model training parameters continue to increase

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 more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Lightweight zero sample learning algorithm framework based on attribute knowledge
  • Lightweight zero sample learning algorithm framework based on attribute knowledge
  • Lightweight zero sample learning algorithm framework based on attribute knowledge

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0038] The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and specific implementation examples.

[0039] In the zero-shot recognition task, there are visible classes and unseen classes; the classes that can be used for training are visible classes, and the classes that cannot be used for training are unseen classes. In regular zero-shot learning tasks, the identification of unseen classes is done in the inference stage. In the generalized zero-shot recognition task, the recognition of seen and unseen classes is done in the inference stage. The present invention assumes that there are U categories in total for unseen categories, and S categories in total for visible categories.

[0040] For the convenience of subsequent operations, U unseen classes are respectively recorded as class 1, class 2... class U, and S visible classes are respectively recorded as class U+1, class U+2,... class U+S.

...

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

No PUM Login to View More

Abstract

The invention discloses a lightweight zero-sample learning algorithm framework based on attribute knowledge. The lightweight zero-sample learning algorithm framework is composed of a feature extraction module, an attribute recognition module and a classifier module. The feature extraction module extracts visual features of pictures, the attribute recognition module performs attribute recognition on each picture and outputs attribute probability, and the classifier module comprises an attribute knowledge graph and adopts a generalized zero sample method and a conventional zero sample method to recognize the pictures based on priori knowledge and attribute probability of the attribute knowledge graph. And outputting an identification result. According to the invention, under the condition of low training cost, high-precision identification is realized; the method is suitable for incremental learning scenes, and the model can retain learned knowledge when learning new knowledge; a feature extraction module in the model framework can be replaced according to precision and computing power requirements; the two-step recognition mechanism can effectively relieve the domain drift problem in generalized zero sample learning.

Description

technical field [0001] The invention belongs to the technical field of image recognition, and in particular relates to a lightweight zero-sample learning algorithm framework based on attribute knowledge. Background technique [0002] Convolutional Neural Network (CNN), as a feedforward neural network, is widely used in various fields such as computer vision and natural language processing. However, with the continuous improvement of the performance of deep learning algorithms, the required labor costs and training costs also continue to increase. First, training a model requires a large dataset. Researchers need to manually collect and label thousands of training samples for each class to make a dataset. This requires a lot of labor costs. Not only that, it is difficult to collect a large number of pictures for each category in reality. Second, as the number of parameters of deep learning models continues to grow, so does the training time. In addition, there is still a...

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
no application Login to View More
IPC IPC(8): G06V10/764G06V10/82G06V10/40G06K9/62G06N3/04G06N3/08
CPCG06V10/765G06V10/82G06V10/40G06N3/08G06N3/047G06N3/045G06F18/2415
Inventor 刘强张泽欢
Owner TIANJIN UNIV