Hi,
I have an image segmentation project for which I saw that the auto-segment tool is not very useful unfortunately. How can I turn off this feature for a given annotation project with the python SDK?
As far as I have seen, the image is already being processed by the auto-segment model as soon as I select the “segmentation” tool, not only when I select the “auto segment: box” tool for example. I would like to completely turn off auto-segment functionality for a given project to save some cost.
Hey @philipp.andermatt ,
Auto-segment is part of the image editor by default and as such we cannot really cherry pick which project it should be working or the opposite.
However, curious on how this is not working for you? Any insight you can provide? We would like to accommodate as much use case as possible!
And we do not charge for the use of Auto segment (in case this was a concern).
Thank you.
PT
Hi @ptancre,
Thank you for your reply!
The segmentations created by the auto-segment tool do not always work very well (medical images). I was indeed thinking that it would save some cost if I could turn off auto-segmentation. Since this is not the case though as you stated, I would like to understand where the model costs that I see in my usage overview come from. I do not use foundry models for prediction and I do not add data rows to model runs. I import pre-labels with lb.MALPredictionImport.create_from_objects(), but since I provide the pre-labels on my own I did not expect this to incur any costs (no model run required). Is my assumption wrong? If yes, how can I import locally produced predictions into labelbox without starting a model run?
The LBU count toward model are (likely) coming from those pre-labels import, the name might be misleading here but we charge as well for import via project.
Many thanks,
PT
Hmm, this is indeed confusing. I checked the detailed usage overview (see image below) and the used LBUs are indeed listed under the model tab. If I import labels as ground truth (with `lb.LabelImport.create_from_objects’), does that also incur cost? If not, why does MAL pre-label import incur costs if I just want to provide my own labels, without actually running a model?
We added this a few week back to acknowledge this :
MAL import creates pre-labels (so not charged as Annotate because not a label) but charges as prediction (which is factor into Model) since it will be render in the project and would be able to curate those via Catalog.
Hope this clarifies.
Okay, thank you for the info. Does the same apply also for annotations that are imported as ground-truth (with lb.LabelImport.create_from_objects)?
So ground-truth, would be charge as Annotate since it would be a label.
1 Like