Questions with merging the detection results.

[Chasel-Tsui]

Hello, I like this well-structured repository for oriented object detection! But, I have several questions when using it.
Q1: I appreciate the automatic detection result merging process on the DOTA. Will the tested AP differs notably between the merged results and the results without merging?
Q2: There is another popular code repository named MMRotate. I am conducting experiments on the OBBDetection and the MMRotate simultaneously. Has anyone tested the performance of the same model (e.g. RetinaNet) on these two repositories? Will the detection results differ greatly in these two repositories?