In this code I found , line labels = torch.ones((records.shape[0],), dtype=torch.int64) ,that there is only one class and 0 in the case of Faster RCNN is reserved for the B