Labeling image segmentation.

[idkwid2022]

Hi,

I want to ask about the code, I don't know where to ask since I don't know if it's bug or just my lack of understanding.

I'm using image segmentation template and on data.py there are lines that I don't understand

code-generator/src/templates/template-vision-segmentation/data.py

Lines 32 to 54 in f137219

	target_names = [
	"background",
	"aeroplane",
	"bicycle",
	"bird",
	"boat",
	"bottle",
	"bus",
	"car",
	"cat",
	"chair",
	"cow",
	"diningtable",
	"dog",
	"horse",
	"motorbike",
	"person",
	"plant",
	"sheep",
	"sofa",
	"train",
	"tv/monitor",
	]

I guessed it's the categories/classes names, but I can't find any reference to the list target_names, even on the parent classes VOCSegmentation, _VOCBase, VisionDataset, and Dataset.

And that brings the next question, how do I label the segmentation result? There are no link between the label and the image dataset right?

Using CVAT, I got labelmap.txt which contains label and mask color, so I was thinking it's the link, but then I can't find any code reading the file.

So, I'm confused. Could anyone please help me with this?

Thank you.

[vfdev-5]

Good question, concerning target_names this image segmentation template was ported from ref example in ignite where we also plot Confusion Matrix : https://github.com/pytorch/ignite/blob/master/examples/references/segmentation/pascal_voc2012/main.py#L156
So yes, it is unused in this template.
Target names is basically a map between integer mask indices and names.

[idkwid2022]

Thanks for clearing that up. I guess I have to read ignite's ref example. 😰