How to label my own dataset for Unet?

[THULiusj]

If I want to use Unet to train my own dataset, how to label it?

[matsujju]

@THULiusj There are many annotation tool you can use according to your need(whether you need to label images using polygon or rectangle shape). You can look here to choose best according to your need. https://github.com/mrgloom/awesome-semantic-segmentation Go to Annotation tool below in the link.

[THULiusj]

Thanks @matsujju ,but if I want to train with Unet, polygon or rectangle is not enough. I have to generate an image with white background and black edge. Not sure how to draw that.

[matsujju]

@THULiusj They will do that. Just read the tutorials before using them according to the format you want.

[THULiusj]

Thanks @matsujju , found it, will explore it!

[ciacci]

@THULiusj what tool did you used to create binary (black and white) images for u-net training? I am interesting in this too, in order to obtain BW labelled images in tif format.

[jizhang02]

hello, how to convert 30 predicted images into one tif image? Thanks.

[cena001plus]

@THULiusj what tool did you used to create binary (black and white) images for u-net training? I am interesting in this too, in order to obtain BW labelled images in tif format.

labelme is good

[Subhajyotidas]

Hi, I tried to use labelme, but it returns json file. How to convert it to binary masking? Or should I try different tool?