I am working with Microscope images of Raptor blood. I need to annotate a large set of images into types of cells and parasites. Red blood cells have 100s of times more occurrences than the other objects(eg Heterophils, Leucocytozoon,…). See example image attached. I have manually labeled a couple of images but need to go through 100s and it takes 1-2 hours each. What is the best way to use automated labeling with label box.
Hey @kcmenke6 ,
You could use a model via Foundry to detect those, we have several vision model that could help out, however given the very specialise topic, you probably need to use a specialised model (I look around and this GitHub - TIO-IKIM/CellViT: CellViT: Vision Transformers for Precise Cell Segmentation and Classification looks promising).
The workflow would be to run the inference in your infra and then send the prediction to Labelbox so you can review and control (either via MAL or Label import).
This would reduce the amount of time you need.
Alternatively (I haven’t tested) but if you only need to label image for a certain use case, for instance the presence of Leucocytozoon you could try to use LLM via a classifier.
Many thanks,
PT
Lance,
Thanks for getting back to me. I will take a look at you gethub reference. I appreciate your pointer. My only experience is with using tenserflow on my pc but i have been avoiding this method.