yolov5-seg

output0 : [1,18900,117]
output1 : [32,160,120]

output0:[1(imgnum),18900(box),[117->[4(xywh),1(boxconf),80(classconf),32(masks全连接权重)]]
output1:[32(depth),160(featuremapH),120(featuremapW)]

output0 -> linear [bboxnum * 32] @ [32, (160,120)] -> upsample [5,(640,480)] -> maskres
output1 -> [1,18900,[4(xywh),1(boxconf),80(classconf)]] -> NMS [1,bboxnum,[4(xywh),1(classconf),1(class)]] -> detres
yolov5-seg_第1张图片

你可能感兴趣的:(AI,segmentation,Computer,Vision,yolo)