consistency of label colors to multiple images

[mwangaz]

I have a question about using labelme to create image labels for training the fully convolutional network (FCN). I find the color of the label depends on the order of creating the polygon: 2 objects will be labelled with red and green color. But what if I have another 2 different objects on an new image? The label color for them will still be red and green, which would affect the label images for training the segmentation model. How can I make sure the label color of the same object on different images is the same? Thanks.

[wkentaro]

I usually create a python script that converts all label images to have the consistent label value.
Maybe I can create a sample in a few weeks if I have time.

[nklein23]

Any news regarding this issue? I am also trying to label multiple images for segmentation but apparently the coloring pattern differentiates from image to image.

[wkentaro]

You can use labelme2voc.py: https://github.com/wkentaro/labelme/blob/master/examples/semantic_segmentation/labelme2voc.py
The output is the consistent colors of labels:

[shivarajkarki]

Thank you, it worked