Thank you for your awesome work! I'm currently using def detr for customized dataset and get a bit confused on the format of target bbox format. According to loss function, the format should be cxcyhw.
But the getitem method of coco dataset is xyxy.
Does the format matter given it is an end-to-end set prediction? And if I want to use pretrained weight on coco dataset, which format should I use?
Thank you for your awesome work! I'm currently using def detr for customized dataset and get a bit confused on the format of target bbox format. According to loss function, the format should be cxcyhw.
But the getitem method of coco dataset is xyxy.
Does the format matter given it is an end-to-end set prediction? And if I want to use pretrained weight on coco dataset, which format should I use?