Caution
Work in progress!
We train YOLO8 models on various datasets to infer cars on orthophotos.
It's still very inaccurate, but have a look:
Note that it is trained only on images of areas somewhere in the range of zoom level 19 and 20. Zoom to some location and select a model.
Thats the output for example of the model self-svd-after-best
(clicking link IMG S
):
Starting annotating some original tiles/img from Vienna could improve the accuracy.
Used trainingsdata is listed under data
.
Preparation scripts are still under scripts/playground/prepareTraining
Model weights