Example for pre-annotating videos with custom model and tracker

[makseq]

We don't have such an example yet, however you can contribute it :-)

1. You can check how to prepare pre-annotations for LS here: https://labelstud.io/guide/predictions.html#Format-pre-annotations-for-Label-Studio

2. As object tracker model you can try Yolo + Deep Sort

3. How to write ML model wrapper for LS: https://labelstud.io/guide/ml_create.html

[ozayr]

I am currently facing the issue where the entire model class is build re-instantiated when ever a predict request is made, I have not dug into the flask code to figure out why. Any idea why this is happening I would assume the class is created once and then the endpoint just calls its predict function. Like this I cannot save object tracks in BaseModel class hence my asking for an example with a custom model and tracker.

[KonstantinKorotaev]

Hi @ozayr Do you use active learning? Please check if you have webhook enabled in project settings

[xiazi-yu]

@ozayr Have you achieved it?

[xiazi-yu]

@makseq The number of image frames decoded by the ML model and Label Studio doesn't match. Is there any way to fix this?