Pixellabeldata is not showing the segmented inforamtion in gTruth file.

I have used Image labeler tool to label images (for pixel segmentation). When I generate gTruth file and load it, I only get pixellabeldata which only contains the address of image location. No other inforamtion like which pixel belongs to which object is mentioned in that gTruth file. In that case how I would be able to train my network using this gTruth file? and how can i generate such gTruth file that contains all the required inforamtion that which pixels belongs to Segment A and which belongs to segment B.?

Answer 1

[Asvin Kumar]

There seems to be two questions to be addressed.

1. It is important to understand the kind of data being exported from the Image Labeller app. You can read about that here. It is expected for the ‘gTruth’ object to only maintain a correspondence to the labelled image stored in the ‘PixelLabelData’ folder. Here’s an example that will help make this clear.
2. Regarding using the labelled data to train a network, you would have to call the appropriate function as mentioned here. Based on what you’ve explained in your question, it seems likely to me that you will need the pixelLabelTrainingData function. You can also check out examples from objectDetectorTrainingData to see if it helps with understanding better.