Bug: Fixed non-latin symbols handling in DXF Georeferencing
AI has become a bit “smarter”
Newly trained detectors are more robust in different light conditions
For annotation, without any training zone, an invisible training zone with the same shape as annotation has been added
That may reduce the number of training faults due to the absence of a training zone. But for better detection results it is still very important to add training zones manually. Training zones help to identify background.
Sometimes on maps, there was experienced some detections outside of the map area
Fixed cases leading to such behavior