Hi, I have some experience in ML training and some years ago I contributed to mapillary object detection.
At the panoramax SOTM EU talk I learned, that there seems to be a labelstudio instance which is used to collect detections in photos. I guess its similar to the panoramax YOLO detection tutorial, the team contributes trainingdata and applies labels as a crowd?
Sorry for this very late answer.
Yes we have a label-studio instance we used to annotated pictures to train the model for face and license plate detection used by our blurring API.
The instance is open on invitation only as we cannot provide a public access to pictures containing faces and license plates that are not blurred.
I published the resulting model and also published other trained model for road sign detection and classification (with their annotated data as these can be public) :
Hi, thanks for the details. Do you think it would be worthwhile that others support you on labeling plates / faces? Not sure if this will increase accuracy e.g. by contributing other countries etc … ?
Is there any interest in extending the object detection for non-privacy related objects? Mapillary for example detects and segments different technical objects of public infrastructure (benchs, hydrants, …). I guess we have a huge and encouraged community to collect photos and create a huge collection of labels 
Regarding plates/faces, the best thing to do is to report the ones with false negatives so that we can add them in the next annotation/training.
For other non privacy related objects, we tarted with an obvious one: road signs.
The models have been trained mostly with pictures coming for France and are very good for the french road signs.
There is no choice made so far about extending this kind of detection on the Panoramax roadmap because the core team is small, and we have a lot of things to do with more basic things like editing pictures, add semantic details to them (some of which could come from computer vision detections), etc.
It is a bit like OSM… should the core of OSM provide a lot of features and services or stay focus on the most important things (the collaborative database, its API and running those service 24/7) and let an open ecosystem create the nice features and services around the core data ?
Right now, we shared our way of using Panoramax pictures with computer vision tools like YOLO, so that if someone wants to build something they have a least the basics to start.
One thing we plan to do is to setup an open label-studio instance.