Sir, i have used a quadrilateral(not special ones like rectangle) to annotate the aerial images and used the same format in which DOTA images were annotated and trained the model with the help of transfer learning..
The confidence score for the prediction is very less (order of 0.001) and the obtained bounding boxes are not covering the entire object..
There is one major problem, my images are aerial view of the seashore,harbor, sea and land near the shore..When the model is predicting, it is showing harbor everywhere in the sea...
My image size is around (2000020000 pixels) and the object is roughly around (3030 pixels)
So, what can be done to improve my results????
Any help will be much appreciated...
Sir, i have used a quadrilateral(not special ones like rectangle) to annotate the aerial images and used the same format in which DOTA images were annotated and trained the model with the help of transfer learning.. The confidence score for the prediction is very less (order of 0.001) and the obtained bounding boxes are not covering the entire object.. There is one major problem, my images are aerial view of the seashore,harbor, sea and land near the shore..When the model is predicting, it is showing harbor everywhere in the sea... My image size is around (2000020000 pixels) and the object is roughly around (3030 pixels) So, what can be done to improve my results???? Any help will be much appreciated...