search

Raj-08 / tensorflow-object-contour-detection

A tensorflow implementation of object-contour-detection with fully convolutional encoder decoder network
MIT License
107 stars 40 forks source link

new labels #7

Open manutom opened 5 years ago

manutom commented 5 years ago

GT[GT!=255]=0 in create_labels.py practically makes all pixels with value 0 which is weird

rubeea commented 4 years ago

Yeah you are right. Any leads on solving the matter so far? I didn't run create_labels.py to generate enhanced labels for training. Instead I trained with the ground-truth labels of PASCAL-VOC dataset.

itziarrr commented 4 years ago

For me it happened because the labels from the dataset have their countours in a yellowish colour, not white. By changing it to GT[GT!=220]=0, I now obtain only the contours. Also added GT[GT==220]=255 to make them white. Hope this helps!

keepgoing365 commented 4 years ago

GT [GT！= 220] = 0 GT [GT == 220] = 255 What do these two lines of code mean?

对我而言，发生这种情况是因为数据集中的标签的计数是淡黄色而不是白色。 通过将其更改为GT [GT！= 220] = 0，我现在仅获得轮廓。 还添加了GT [GT == 220] = 255使它们变为白色。 希望这可以帮助！

itziarrr commented 4 years ago

220 in my case was the intensity of the pixels of the contours. Therefore, GT [GT！= 220] = 0 means 'make all that is not contour black' (pixel level = 0) And GT [GT == 220] = 255 makes the contours truly white.