在compute_pseudo_label_loss()的时候已经对teacher生成的伪标签做了坐标变换，为什么在unsup_rcnn_cls_loss()的时候还要变换一次呢？

[wzr0108]

这里是对teacher生成的伪标签做了坐标变换吧？

def compute_pseudo_label_loss(self, student_info, teacher_info):
    M = self._get_trans_mat(
        teacher_info["transform_matrix"], student_info["transform_matrix"]
    )
    pseudo_bboxes = self._transform_bbox(
        teacher_info["det_bboxes"],
        M,
        [meta["img_shape"] for meta in student_info["img_metas"]],
    )

然后在unsup_rcnn_cls_loss()又做了一次坐标变换

M = self._get_trans_mat(student_transMat, teacher_transMat)
aligned_proposals = self._transform_bbox(
    selected_bboxes,
    M,
    [meta["img_shape"] for meta in teacher_img_metas],
)