首页|Patent Issued for Learning semantic segmentation models in the absence of a port ion of class labels (USPTO 12118788)
Patent Issued for Learning semantic segmentation models in the absence of a port ion of class labels (USPTO 12118788)
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
Reporters obtained the following quote from the background information supplied by the inventors: “Sematic segmentation refers to a computer vision task in whic h regions of an image are labeled based on the items that are included in the im age. The semantic segmentation task is to label each pixel of the image with a c orresponding class of what is being represented by that pixel. “Deep learning approaches to computer vision require large amounts of labeled an d carefully curated data to perform well. Semantic segmentation, where each pixe l in an input image is classified as belonging to a semantic class is particular ly demanding, as each pixel in each training example must be labeled to enable s uccessful model training. As a result, while semantic segmentation is an importa nt component in a variety of tasks, from scene analysis to self-driving cars, it is dependent on the availability of expensive labeled data.”
NETL
NSTL
