Automate labeling with your models

With Picsellia, you can automatically annotate a Dataset with a trained model, you can use every models in the Model Hub or in your organization hub.

If you don't have any trained model yet learn how to upload a trained model πŸ‘‰ here

Select the Dataset that you want to pre-annotate

You must have some labels configured, if you don't know how to do this, learn about it πŸ‘‰ here​

Once your OK with this let's go to annotations -> automate in your dataset page

Select the labels for your pre-annotation

You can chose to automate labeling for the number of classes that you want.

Once you have selected the classes to be pre-annotated you can click on next

Chose the model to use

You can chose the model that you want to use from the public hub, (mostly coco classes with different architectures) or in your own model hub.

Select what class to detect and pre-annotate as your labels

Sometimes it's valuable to merge some predictions into one class, you can simply do it by selecting multiple classes to be annotated as your selected labels.

Select confidence treshold for each classes

Once you are set up, you need to select the treshold of each predictions to be accepted ( default is 0.5 ) use this with caution, a small confidence threshold will give lot of predictions but the quality will be questionable :)

Then press execute instructions and go grab a coffee :) β˜•οΈ

Your work here is done πŸš€