Semantic Segmentation | semantic segmentation state of the art 2019
Semanticsegmentation,orimagesegmentation,isthetaskofclusteringpartsofanimagetogetherwhichbelongtothesameobjectclass.Itisaformofpixel-levelpredictionbecauseeachpixelinanimageisclassifiedaccordingtoacategory.SomeexamplebenchmarksforthistaskareCityscapes,PASCALVOCandADE20K.ModelsareusuallyevaluatedwiththeMeanIntersection-Over-Union(MeanIoU)andPixelAccuracymetrics.(Imagecredit:CSAILVision[1])ReferencesCSAILVision(github.com)
Semantic segmentation, or image segmentation, is the task of clustering parts of an image together which belong to the same object class. It is a form of pixel-level prediction because each pixel in an image is classified according to a category. Some example benchmarks for this task are Cityscapes, PASCAL VOC and ADE20K. Models are usually evaluated with the Mean Intersection-Over-Union (Mean IoU) and Pixel Accuracy metrics.
( Image credit: CSAILVision[1] )
References CSAILVision (github.com)