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

Multi-scale edge detection method under deep supervision

An edge detection and multi-scale technology, applied in the field of multi-scale edge detection, can solve problems such as difficult optimization, invalid modeling of long-distance correlation, and difficulty in transmitting information at long-distance locations, and achieve the effect of accurate and reliable edge detection

Pending Publication Date: 2021-03-30
JIANGNAN UNIV
View PDF0 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, direct repetition of convolutional layers is computationally inefficient, difficult to optimize, and difficult to transmit information between distant locations, which leads to ineffective modeling of long-range correlations

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
  • Multi-scale edge detection method under deep supervision
  • Multi-scale edge detection method under deep supervision
  • Multi-scale edge detection method under deep supervision

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0067] We implement our network using PyTorch which is well known in the industry. Initialize our network with VGG16 pre-trained on ImageNet. The threshold λ for calculating the loss in the BSDS500 and NYUD datasets is set to 1.1 and 1.2, respectively.

[0068] The SGD optimizer randomly extracts 10 images in each iteration, and the global learning rate is set to 1e-6, which is divided by 10 after every 10K iterations. Momentum and weight decay are set to 0.9 and 0.0002, respectively. We did 40K iterations in total. All experiments of the present invention are carried out on NVIDIA 1080GPU.

[0069] We tested the edge detection performance under common evaluation metrics, Optimal Dataset Scale (ODS), Optimal Image Scale (OIS) and Average Precision (AP). Before evaluation, we use non-maximum suppression (NMS) to refine edges, e.g. Based on previous work, the localization tolerance of the maximum allowed distance between the predicted edge and the ground truth for the BSDS5...

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 multi-scale edge detection method under deep supervision. According to the method, local features and corresponding global correlation can be combined, channel response is self-adaptively recalibrated, a network is guided to ignore irrelevant information, and correlation between related features is emphasized. A series of ablation experiments are carried out on the methodon a BSDS500 data set and an NYUD data set, and the effectiveness of a multi-scale deep supervision self-attention module algorithm is proved. Compared with other most advanced edge detection networks, the algorithm has better performance, the prediction precision is improved by using fewer parameters, and the score of an ODS measurement value of 0.815 is realized on a BSDS00 data set and is 0.9%higher than that of other existing algorithms.

Description

technical field [0001] The invention belongs to the field of edge detection, and in particular relates to a multi-scale edge detection method under deep supervision. Background technique [0002] Edge detection aims to extract object boundaries and visually distinct edges in natural images, which are important for advanced computer vision tasks such as image segmentation, object detection / recognition. Edge detection has a rich history as a foundation for advanced tasks, and we now focus on a few representative works that have proven to be of great significance. Early traditional methods include the Sobel detector, zero-crossing detection, and the widely used Canny detector. Pb, gPb, Sketch token, and Structured Edges use sophisticated learning paradigms to distinguish edge pixels based on handcrafted features such as brightness, color, gradient, and texture. However, it is difficult to represent semantic meaning by applying low-level visual cues. [0003] The edges of ima...

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
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/46G06K9/62G06N3/04G06N3/08G06T3/40
CPCG06T3/4007G06N3/08G06V10/44G06N3/047G06N3/048G06N3/045G06F18/2415
Inventor 孙俊张旺吴豪吴小俊方伟陈祺东李超游琪冒钟杰
Owner JIANGNAN UNIV