Should I delete old frigate+ images if re-training from scratch?

[ics460]

I recently upgraded to v0.17.0 and have started to experiment with YOLOv9 on Coral as well as LPR and Face Recognition.

One of my cameras has 600+ images trained in frigate+, however the objects trained in those images were limited at the time (didn't include face, license plate, some others). Additionally, I was not following all of the best practices (#21374) and probably fed in some data that is less than ideal.

It seems like a good chance to start from scratch, though that would require deleting all of my images in frigate+ which seemingly can't be done in bulk. While it appears that I can give the camera a new name and submit images to frigate+ under the new identifier, I am wondering if the images under the old camera would affect the model used by the new camera?

[NickM-27]

It's not recommended, the images are still helpful training data

[blakeblackshear]

Happy to help cleanup old images if you want. Just send me an email at blake@frigate.video.

The images in Frigate+ aren't connected to any particular model type. It's a general purpose training set that can be used to train any model type supported now or in the future.