TRI-POSE-Net: Adaptive 3D human pose estimation through selective kernel networks and self-supervision with trifocal tensors DOI Creative Commons
Nabeel Ahmed Khan, Aisha Ahmed Alarfaj, Ebtisam Alabdulqader

et al.

PLoS ONE, Journal Year: 2024, Volume and Issue: 19(12), P. e0310831 - e0310831

Published: Dec. 5, 2024

Accurate and flexible 3D pose estimation for virtual entities is a strenuous task in computer vision applications. Conventional methods struggle to capture realistic movements; thus, creative solutions that can handle the complexities of genuine avatar interactions dynamic environments are imperative. In order tackle problem precise estimation, this work introduces TRI-POSE-Net, model intended scenarios with limited supervision. The proposed technique, which based on ResNet-50 includes integrated Selective Kernel Network (SKNet) blocks, has proven be efficient feature extraction customised specifically scenarios. Furthermore, trifocal tensors their trio-view geometry allow us generate ground truth poses from 2D poses, resulting more refined triangulations. Through approach, estimated single RGB image. Moreover, approach was evaluated HumanEva-I dataset yielding Mean-Per-Joint-Position-Error (MPJPE) 47.6 under self-supervision an MPJPE 29.9 full comparison other works, performed well paradigm.

Language: Английский

Research on Change Detection Algorithm for Remote Sensing Images Based on the Attention-Guided Multiscale Feature Fusion DOI

正正 贾

Computer Science and Application, Journal Year: 2025, Volume and Issue: 15(02), P. 1 - 12

Published: Jan. 1, 2025

Language: Английский

Citations

0
Comparison of Preprocessing Methods Impact on Detection of Soldering Splashes Using Different YOLOv8 Versions DOI Creative Commons
Peter Klčo, Dušan Koniar, Libor Hargaš

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 5, 2024

Abstract Quality inspection of electronic boards during manufacturing process is crucial step especially in the case specific and expensive power electronics modules. Soldering splashes occurrence decreases reliability electric properties final products. The aim this paper to compare different YOLOv8 models (small, medium, large) with combination basic image preprocessing techniques achieve best possible performance designed algorithm. As methods contrast limited adaptive histogram equalization (CLAHE) color channels manipulation are used. Results show that suitable model leads increase recall parameter. In our task, can be considered as most important metric. results supported by standard two-way ANOVA test.

Language: Английский

Citations

0
Identification and characterisation of Type Ⅱ MnS via YOLO deep learning method DOI

Qiu‐wei Zheng,

Xiaoyong Gao, Lifeng Zhang

et al.

Ironmaking & Steelmaking Processes Products and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 14, 2024

In order to solve the problem that current inclusion characterisation methods can only identify single-point inclusions, identification and model of type II MnS inclusions was established, related software developed. Scanning electron microscopy used for establish a database MnS. You Only Look Once deep learning realise recognition with rate 0.01. Image post-processing technologies such as edge detection grey value extraction were characterise identified MnS, accuracy confirmed by visualising information. This method achieved up 91.485% mAP 0.5 in identifying Type recall 85.924% 83.333%, respectively. The could be accurately completed adjusting threshold. detect greatly saved time reduced error.

Language: Английский

Citations

0
Comparison of Preprocessing Method Impact on the Detection of Soldering Splashes Using Different YOLOv8 Versions DOI Creative Commons
Peter Klčo, Dušan Koniar, Libor Hargaš

et al.

Computation, Journal Year: 2024, Volume and Issue: 12(11), P. 225 - 225

Published: Nov. 12, 2024

Quality inspection of electronic boards during the manufacturing process is a crucial step, especially in case specific and expensive power modules. Soldering splash occurrence decreases reliability electric properties final products. This paper aims to compare different YOLOv8 models (small, medium, large) with combination basic image preprocessing techniques achieve best possible performance designed algorithm. As methods, contrast-limited adaptive histogram equalization (CLAHE) color channel manipulation are used. The results show that suitable model methods leads an increase recall parameter. In our task, can be considered most important metric. supported by standard two-way ANOVA test.

Language: Английский

Citations

0
TRI-POSE-Net: Adaptive 3D human pose estimation through selective kernel networks and self-supervision with trifocal tensors DOI Creative Commons
Nabeel Ahmed Khan, Aisha Ahmed Alarfaj, Ebtisam Alabdulqader

et al.

PLoS ONE, Journal Year: 2024, Volume and Issue: 19(12), P. e0310831 - e0310831

Published: Dec. 5, 2024

Accurate and flexible 3D pose estimation for virtual entities is a strenuous task in computer vision applications. Conventional methods struggle to capture realistic movements; thus, creative solutions that can handle the complexities of genuine avatar interactions dynamic environments are imperative. In order tackle problem precise estimation, this work introduces TRI-POSE-Net, model intended scenarios with limited supervision. The proposed technique, which based on ResNet-50 includes integrated Selective Kernel Network (SKNet) blocks, has proven be efficient feature extraction customised specifically scenarios. Furthermore, trifocal tensors their trio-view geometry allow us generate ground truth poses from 2D poses, resulting more refined triangulations. Through approach, estimated single RGB image. Moreover, approach was evaluated HumanEva-I dataset yielding Mean-Per-Joint-Position-Error (MPJPE) 47.6 under self-supervision an MPJPE 29.9 full comparison other works, performed well paradigm.

Language: Английский

Citations

0