liu.seSearch for publications in DiVA
3334353637383936 of 48
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Exploring the applicability of Graph Attention Networks in computer vision and their hardware acceleration
Linköping University, Faculty of Science & Engineering. Linköping University, Department of Electrical Engineering, Electronics and Computer Engineering.ORCID iD: 0000-0003-4870-2768
Linköping University, Department of Electrical Engineering, Electronics and Computer Engineering. Linköping University, Faculty of Science & Engineering.ORCID iD: 0000-0002-5153-5481
2025 (English)In: AccML papers 2025, 2025, article id 3Conference paper, Published paper (Other academic)
Abstract [en]

Edge detection is a fundamental task in computer vision, crucial for object recognition, segmentation, and scene understanding. Traditional methods often fail to capture complex edge structures due to their inability to model intricate relationships between pixels. Graph Neural Networks (GNNs), particularly Graph Attention Networks (GATs), have shown promise in addressing these limitations by leveraging graph structures to model pixel relationships. This paper explores the applicability of Graph Attention Networks in edge detection, highlighting their advantages over ordinary Graph convolutional Networks (GCNs) through rigorous mathematical reasoning. We integrate GATs into an edge detection framework based on an encoder-decoder structure with U-Net architecture and provide detailed theoretical and implementation insights. Furthermore, we discuss the hardware acceleration of GCNs and GATs with a reconfigurable dataflow architecture integrated in the Pytorch framework. The experimental results demonstrate the superior performance of GAT-based edge detection and the potential acceleration possible on reconfigurable edge platforms with limited resources. The key advantage of our proposed method is its hardware-friendly design, making it highly suitable for FPGA acceleration while also enabling efficient optimization through pruning of the network. 

Place, publisher, year, edition, pages
2025. article id 3
Keywords [en]
edge detection, GNN, Graph Attention Networks, encoder-decoder structure, U-Net
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:liu:diva-220183OAI: oai:DiVA.org:liu-220183DiVA, id: diva2:2023448
Conference
7th Workshop on Accelerated Machine Learning (AccML) on HiPEAC 2025 Conference, 21st January, 2025, Barcelona, Spain
Available from: 2025-12-19 Created: 2025-12-19 Last updated: 2025-12-19

Open Access in DiVA

fulltext(871 kB)47 downloads
File information
File name FULLTEXT01.pdfFile size 871 kBChecksum SHA-512
4caf91d057343b0a8dc7655ca0bca0bfbb9ea54b6c69a4487363989dfa72126fbb9127d3660e2e70508f2eb8e0571c3eb052c34c784a919e7b38f11c90fad2ba
Type fulltextMimetype application/pdf

Other links

Fulltext from conference

Authority records

Khalili Sadaghiani, AbdolvahabNunez-Yanez, Jose

Search in DiVA

By author/editor
Khalili Sadaghiani, AbdolvahabNunez-Yanez, Jose
By organisation
Faculty of Science & EngineeringElectronics and Computer Engineering
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 759 hits
3334353637383936 of 48
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf