search

HumanSignal / label-studio

Label Studio is a multi-type data labeling and annotation tool with standardized output format
https://labelstud.io
Apache License 2.0
19.26k stars 2.39k forks source link

Large TIFF Images such as Histopathology images #317

Open mcombalia opened 4 years ago

mcombalia commented 4 years ago

Is it possible to upload large tiff images such as histopathology images and have them labeled for segmentation?

deppp commented 4 years ago

It's possible as long as you can convert them into jpg and the browser supports the size of it. Is there any reason why you would want to label the original tiffs?

mcombalia commented 4 years ago

Thanks for answer.

These images are extremely high dimensional and usually transforming them to JPG is not adequate, since they are stored in a pyramidal form in order to save space (see example: https://brd.nci.nih.gov/brd/specimen/GTEX-1117F-0126)

Please zoom the image.

These images are used a lot in the medical community.

HenryNebula commented 4 years ago

Same feature needed here. Seems like OpenSeadragon is a good frontend package for that task. Whole slide images seem to be loaded using ajax-like methods, so it doesn't require loading the entire image to the browser at once. I am not an expert in frontend development and I am having trouble integrating this feature to existing framework. Would appreciate if any help can be provided with this issue! Thanks!

jrdalenberg commented 3 years ago

@mcombalia why not feed all the images of the lowest layer of the pyramid for labeling? I do not think labeling the whole image in a single go is very user friendly nor is it practical for object classification. You can map the results back to the original image.

I am working with 1-50GB TIFF images and do it like that.

mcombalia commented 3 years ago

@jrdalenberg while that is an option, this is not useful when researchers need to zoom to image regions in order to label them. An option is to do patches but then you are losing the contextual information for labeling. The platform that I am using ATM for that is Digital Slide Archive.