首页|New Intelligence Technology Findings from Harbin University Described (Local Sal iency Consistency-based Label Inference for Weakly Supervised Salient Object Det ection Using Scribble Annotations)
New Intelligence Technology Findings from Harbin University Described (Local Sal iency Consistency-based Label Inference for Weakly Supervised Salient Object Det ection Using Scribble Annotations)
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 DailyNews Daily News-Fresh data on Machine Learning - Intel ligence Technology are presented in a newreport. According to news reporting fr om Harbin, People's Republic of China, by NewsRx journalists,research stated, " Recently, weak supervision has received growing attention in the field of salien t objectdetection due to the convenience of labelling. However, there is a larg e performance gap between weaklysupervised and fully supervised salient object detectors because the scribble annotation can only providevery limited foregrou nd/background information."The news correspondents obtained a quote from the research from Harbin Universit y, "Therefore, anintuitive idea is to infer annotations that cover more complet e object and background regions for training.To this end, a label inference str ategy is proposed based on the assumption that pixels with similar coloursand c lose positions should have consistent labels. Specifically, k-means clustering a lgorithm was firstperformed on both colours and coordinates of original annotat ions, and then assigned the same labels topoints having similar colours with co lour cluster centres and near coordinate cluster centres. Next, thesame annotat ions for pixels with similar colours within each kernel neighbourhood was set fu rther."
HarbinPeople's Republic of ChinaAsiaIntelligence TechnologyMachine LearningHarbin University
NETL
NSTL
