I’m creating machine learning model on orthomosaic map imported from the drone Nakesha so is there any trick to improve the accuracy of train machine learning model ?
2 Likes
Hi @bunny the trick is to select appropriate train and test areas . You need to make sure the annotated areas are having enough good objects and clear features of them .
Hello @bunny
When you are annotating object it should cover most of the examples include variety what i mean by this is suppose i am making a detector of a car i would annotate various cars which includes different color car’s , different shapes of car . Providing such examples helps your model to learn your object with higher precision .The higher the number of examples for the model to learn, the higher the accuracy.
Another method to improve your accuracy .
Post Training method :Here’s one of the trick to improve your model’s accuracy feed the detection’s as annotations and re-train your model to increase accuracy. In simple words
suppose you have created a detector for car i would run detection after it considering in your case detection on geo tif , Now click on export geojson and import it in new training task you just saved manual annotation work and edit your detection’s delete wrong detection’s add in a few examples of object which were not detected draw areas and start a training task .
1 Like
Bro its DroneNaksha 