What I want is to detect (new) buildings from a aerial photo. I have a 7.5 cm resolution aerial photo.
Furthermore, I already have a large dataset with polygons of buidings (around 150.000 buildings).
Can I use this dataset to create an annotation file for the positive samples to use in the RasterObjectDetector?
And can I also create negative samples by taken regions from the aerial photo where no buildings are located?
Hi @stefanh,
As far as I know, if you are training your own model using your data, creating the annotation file is a bit of a tedious manual step of drawing rectangles over target objects. If you can find a pre-trained model on buildings from an aerial view, that could potentially save you a lot of time. However something to keep in mind is that the accuracy of the detection results will depend on how well the model is trained, the types of samples used, and how many samples were used.
Otherwise, yes, you can absolutely use the datasets that you mentioned for the samples (positives and negatives respectively).
This blog article from Dmritri contains a successful workflow example that might be a helpful resource for your project. Hope that helps a bit!