Graph fusion network for multi-oriented object detection

Zhang, Shi-Xue; Zhu, Xiaobin; Hou, Jie-Bo; Yin, Xu-Cheng

doi:10.1007/s10489-022-03396-5

Cited by 12 publications

(5 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Connected Component (CC) based methods [21], [22], [23], [1], [24], [2] usually detect individual text parts or characters firstly, followed by a link or group post-processing procedure for generating final texts. SegLink++ [23] use instance-aware component grouping with minimum spanning tree to achieve arbitrary shape text detection.…”

Section: B Connected Component Based Methodsmentioning

confidence: 99%

Arbitrary Shape Text Detection via Boundary Transformer

Zhang¹,

Zhu²,

Yuan³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Arbitrary shape text detection is a challenging task due to its complexity and variety, e.g., various scales, random rotations, and curve shapes. In this paper, we propose an arbitrary shape text detector with a boundary transformer, which can accurately and directly locate text boundaries without any post-processing. Our method mainly consists of a boundary proposal module and an iteratively optimized boundary transformer module. The boundary proposal module consisting of multi-layer dilated convolutions will compute important prior information (including classification map, distance field, and direction field) for generating coarse boundary proposals meanwhile guiding the optimization of boundary transformer. The boundary transformer module adopts an encoder-decoder structure, in which the encoder is constructed by multi-layer transformer blocks with residual connection while the decoder is a simple multilayer perceptron network (MLP). Under the guidance of prior information, the boundary transformer module will gradually refine the coarse boundary proposals via boundary deformation in an iterative manner. Furthermore, we propose a novel boundary energy loss (BEL) which introduces an energy minimization constraint and an energy monotonically decreasing constraint for every boundary optimization step. Extensive experiments on publicly available and challenging datasets demonstrate the stateof-the-art performance and promising efficiency of our method.

show abstract

Section: B Connected Component Based Methodsmentioning

confidence: 99%

Arbitrary Shape Text Detection via Boundary Transformer

Zhang¹,

Zhu²,

Yuan³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…The classic texts in typical scenes and landscapes are more diverse and unpredictable in direct comparison to structured graphical texts [59,61]. The successful extraction of classic texts from these possible kinds of emotional scenes typically using portable robots for accurate navigation, graciously according to traffic signs, the direct detection and critical recognition of license plates, object recognition, and so on are examples of novel applications of automatic text recognition [62].…”

Section: Case Studymentioning

confidence: 99%

Big Image Data 15Vs Model for Intelligent Data Ocean of Multimedia Things

Mahabadi,

Abdolazimi

2024

Preprint

View full text Add to dashboard Cite

The classification of large image data using traditional methods has been based on changes in image gray features, extraction of edge and contour feature information, or conversion between image coordinate sets. However, due to the growth of new image size and big image data (BID) in real-time multimedia communication and future online big data (BD) applications, these methods have become increasingly complex and resulting in poor real-time performance. These complex methods suffer from complex algorithms, massive data communication, slow processing speed, unintelligent predictive modeling, weak data classification, limited accuracy, uncleaned usage data, combined destructive artificial and natural noise, exchanging data values over time, and exhausting updating data for accurate operation of media storage. In the era of the Internet of Multimedia Things (IoMT), most modern devices can accurately capture vast amounts of observational data in the form of valuable images, text, and acoustic recordings. This reliable data must be kept valid for local data processing at different times and for globally extracting knowledge from information at various data levels to address forthcoming challenges related to big image data. The challenges of secure communication require accurate, reliable, fast, and effective information gathering for local decision-making and global knowledge mining of BID to support cyber-physical systems for speedy data virtualization of smart city devices. Complex and uncontrollable problems persist on the outskirts of grids, fogs, and cloud networks regarding data cleansing and privacy protection for intelligent data ocean management. This study developed a new 15Vs model empirically to examine distributed big image data processing based on a modern BID connectivity approach. The model accurately extracts textured features from visible data images and overcomes a unique set of key challenges. The model introduces a strategic analyzer, uses intelligent local agents, and recruits clever global bots to provide a suitable platform for generous support of bot-oriented BD processing. This instantly forms a hierarchical data level for better data acquisition and cleansing, safe privacy protection, suitable information diffusion, and speedy knowledge extraction. Our study presents a novel definition of BID to address the ultimate challenges of data management in a high-risk environment for further big image data modeling research needs.

show abstract

“…Overall, deep learning-based object detection algorithms have developed rapidly and achieved good results in both scientific research and industrial applications [42], [19], [38], [40], [31]. However, these common object detection algorithms are mainly designed for optical natural images rather than underwater images, and their direct application to underwater images often results in poor performance, so special designs for underwater object detection are still needed.…”

Section: Related Workmentioning

confidence: 99%

Underwater Image Object Detection based on Multi-scale Feature Fusion

Yang,

Zhang,

Jiang

et al. 2024

Preprint

View full text Add to dashboard Cite

Underwater object detection and classification technology is one of the most important ways for humans to explore the oceans. However, existing methods are still insufficient in terms of accuracy and speed, and have poor detection performance for small objects such as fish. In this paper, we propose a multi-scale aggregation enhanced (MAE-FPN) object detection method based on the feature pyramid network, including the multi-scale convolutional calibration module (MCCM) and the feature calibration distribution module (FCDM). First, we design the MCCM module, which can adaptively extract feature information from objects at different scales. Then, we built the FCDM structure to make the multi-scale information fusion more appropriate and to alleviate the problem of missing features from small objects. Finally, we construct the Fish Segmentation and Detection (FSD) dataset by fusing multiple data augmentation methods, which enriches the data resources for underwater object detection and solves the problem of limited training resources for deep learning. We conduct experiments on FSD and public datasets, and the results show that the proposed MAE-FPN network significantly improves the detection performance of underwater objects, especially small objects.

show abstract

Graph fusion network for multi-oriented object detection

Cited by 12 publications

References 64 publications

Arbitrary Shape Text Detection via Boundary Transformer

Arbitrary Shape Text Detection via Boundary Transformer

Big Image Data 15Vs Model for Intelligent Data Ocean of Multimedia Things

Underwater Image Object Detection based on Multi-scale Feature Fusion

Contact Info

Product

Resources

About