A method and device for pixel segmentation of feature map rows
A pixel segment and feature map technology, applied in the field of image processing, can solve the problem that the number of deep learning accelerator multipliers cannot meet the needs of convolution calculations
Active Publication Date: 2021-02-23
DEEPBLUE TECH (SHANGHAI) CO LTD
View PDF14 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0006] The present invention provides a method and device for segmenting pixels in a feature map row to solve the problem in the prior art that the number of deep learning accelerator multipliers cannot meet the needs of convolution calculations
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
[0040] With the development of artificial intelligence technology, deep learning has become an important development direction in artificial intelligence recognition technology due to its excellent performance in image recognition.
[0041] The concept of deep learning originally originated from the research of artificial neural networks. For example, a multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning combines low-level features to form more abstract high-level representation attribute categories or features to discover distributed feature representations of data.
[0042] Secondly, deep learning is a new field in machine learning research. It is a method based on representational learning of data in machine learning. Interpret the data.
[0043] In the existing deep learning application process, the convolution calculation of deep learning is based on GUP. When deep learning is performed based on GUP, a deep learning accelerator...
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 discloses a method and a device for segmenting pixels of a feature map row, which are used to solve the problem in the prior art that the number of deep learning accelerator multipliers cannot meet the requirements of convolution calculation. In the embodiment of the present invention, it is judged whether the first number of pixels in the row direction of the feature map is greater than the second number of pixels processed by the deep learning accelerator; Perform segmentation processing to obtain a plurality of pixel segments, wherein the number of pixels included in the pixel segment is not greater than the second number of pixels processed by the deep learning accelerator; perform convolution calculation on each pixel segment. In this way, the pixels in the row direction of the feature map are segmented, and the number of pixels included in the segmented pixel segments is guaranteed to be no greater than the second number of pixels processed by the deep learning accelerator, and then the segmented pixel segments are divided into Carry out convolution calculation, and then complete the convolution calculation for the feature map.
Description
technical field [0001] The present invention relates to the technical field of image processing, in particular to a method and device for pixel segmentation of feature map rows. Background technique [0002] The concept of deep learning originated from the study of artificial neural networks. With the deepening of deep learning research, the amount of computation of convolutional neural networks has changed greatly. In order to meet the needs of computing power in deep learning, the following methods are generally used: (1) through ASIC (Application Specific Integrated Circuits, Application-specific integrated circuit) to increase the rate of deep learning; (2) to increase the rate of deep learning through FPGA (Field-Programmable Gate Array, Field Programmable Gate Array); (3) to improve the rate of deep learning through GPU (Graphics Processing Unit, graphics processing unit) rate. [0003] When the calculation speed of the convolutional neural network is improved throug...
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 Patents(China)
IPC IPC(8): G06T7/11
CPCG06T2207/20084G06T7/11
Inventor 陈海波
Owner DEEPBLUE TECH (SHANGHAI) CO LTD