Context I tried to create my own satellite imagery using computer generated airplanes and bounding boxes so that I could train a ResNet object detection model. It works well with certain types of planes but not others. I'll probably expand the database later to include things like helicopters, fighters, etc. Here are some initial results of my model using this data: ![enter image description here][1] Content Two folders of 500 computer generated satellite images of planes accompanied with an xml of the bounding box coordinates for each. I used VTK to place a 3d model of cargo planes over random satellite imagery from Mapbox. I randomized the location, orientation, pitch, yaw, roll as well as the light position and color to add some variability. Since only one type of plane was used, the object detection model I ended up training could recognize planes similar to it but sometimes not other airliners. I plan on adding helicopters and other types of aircraft in the future. [1]: https://i.imgur.com/dneUfp5.png