How can I load segmentation results from algorithm and then use this tool to finetune ?

[JarveeLee]

No description provided.

[dimimal]

I have this question too. Is it possible to load a predefined mask as an overlay layer and then fine tune the segmentation mask? Ideally, the mask should be an image file.

[dimimal]

@JarveeLee You should name the mask image in the appropriate form in order to load it.

[galexioza]

Same question here. I am wanting to label hundreds of images that look almost exactly the same or at least have the same general layout (e.g. individual plants in an orchard where the sky is always at the top, the plant is always in the middle and the ground is always at the bottom). So it would be good if for each image I am segmenting, it could load the previous image's segmentation results automatically, overlay it, then let me fine-tune it, then load the next image with the previous image's mask overlayed, let me, fine-tune it, and repeat. Alternatively it could use a 'master' mask. The point is that 98% of the image can reuse the mask, there are usually only minor changes that need to be made and reslecting every component each time is far too time consuming.

[YuLin-Cheng]

Hi @dimimal
do you mean just give predefined mask a name like "_watershed_mask.png" or "_mask.png" ?
I have several binary(black and white) images, but in the tool it shows up more colors than only black and white.

Any idea would be appreciated

[daandres]

For loading from a previous work (network, or other), you have to convert your files to PixelAnnotationTools file format.
It is quite simple. Just save your network outputs to a grayscale image with the corresponding label id at each pixel position. And use PNG format.

PixelAnnotationTool does not load color format, those are just saved for better visualization in file browser.

@YuLin-Cheng Maybe you had involved a JPG in your pre processing steps? This yould introduced some compression artifacts.
Can you post a sample of your observations?

[abreheret]

Thank you for @daandres's answer, I add that : the xx_mask.png is gray image. It contains the "R/G/B" with "id/id/id" (define in the config.json) of labels marker mask to generate the xx_watershed_mask.png and the xx_color_mask.png. So if the annotation is more or less the same in a series of images copy the file xx_mask.png to new yy_mask.png to accelerate your annotation.