首页|Astana IT University Reports Findings in Artificial Intelligence (Using syntheti c dataset for semantic segmentation of the human body in the problem of extracti ng anthropometric data)
Astana IT University Reports Findings in Artificial Intelligence (Using syntheti c dataset for semantic segmentation of the human body in the problem of extracti ng anthropometric data)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News – New research on Machine Learning - Art ificial Intelligence is the subject of a report. According to news reporting ori ginating from Astana, Kazakhstan, by NewsRx correspondents, research stated, “Th e COVID-19 pandemic highlighted the need for accurate virtual sizing in e-commer ce to reduce returns and waste. Existing methods for extracting anthropometric d ata from images have limitations.” Our news editors obtained a quote from the research from Astana IT University, “ This study aims to develop a semantic segmentation model trained on synthetic da ta that can accurately determine body shape from real images, accounting for clo thing. A synthetic dataset of over 22,000 images was created using NVIDIA Omnive rse Replicator, featuring human models in various poses, clothing, and environme nts. Popular CNN architectures (U-Net, SegNet, DeepLabV3, PSPNet) with different backbones were trained on this dataset for semantic segmentation. Models were e valuated on accuracy, precision, recall, and IoU metrics. The best performing mo del was tested on real human subjects and compared to actual measurements. U-Net with EfficientNet backbone showed the best performance, with 99.83% training accuracy and 0.977 IoU score. When tested on real images, it accurately segmented body shape while accounting for clothing. Comparison with actual meas urements on 9 subjects showed average deviations of -0.24 cm for neck, -0.1 cm f or shoulder, 1.15 cm for chest, -0.22 cm for thallium, and 0.17 cm for hip measu rements. The synthetic dataset and trained models enable accurate extraction of anthropometric data from real images while accounting for clothing. This approac h has significant potential for improving virtual fitting and reducing returns i n e-commerce.”
KazakhstanAsiaArtificial Int elligenceMachine Learning
NETL
NSTL
