how to run tta for rotated object detection?

[young169]

I add some variables just as https://github.com/open-mmlab/mmyolo/blob/main/configs/_base_/det_p5_tta.py said, and when testing on my own rotated object detections datasets, the dataloader gives me some shape error like " The boxes dimension must >= 2 and the length of the last dimension must be 4, but got boxes with shape torch.Size([2, 8])."
I myself consider it's mainly because of the rotated objects have four points' information while common datasets COCO or PascalVOC have xywh format. So how to fix that?