【扒代码】.repeat_interleave(self.num_objects=3)

            bboxes = torch.cat([
                torch.arange(
                    bs, requires_grad=False
                ).to(bboxes.device).repeat_interleave(self.num_objects).reshape(-1, 1),
                bboxes.flatten(0, 1),
            ], dim=1)
            # print(bboxes)

tensor([[ 0.0000, -0.6644,  0.8709, -0.7267, -2.2695],
        [ 0.0000,  1.4619, -1.0581, -1.8898,  0.7876],
        [ 0.0000, -1.6069, -1.3537,  0.5698,  0.6994],
        [ 1.0000,  0.2495, -0.5027, -1.5294,  0.0376],
        [ 1.0000, -0.8491,  0.4727,  0.4423,  0.5050],
        [ 1.0000,  0.3200, -0.1111, -1.0900, -2.4595],
        [ 2.0000, -0.7400, -0.2743,  1.6047, -0.5900],
        [ 2.0000,  0.3749, -0.0864,  1.2539, -0.9245],
        [ 2.0000, -0.1543, -1.1329,  1.7370,  0.7067],
        [ 3.0000,  1.6104, -1.0426,  0.8122, -0.1253],
        [ 3.0000, -0.6058, -0.5058, -1.2164, -0.7558],
        [ 3.0000, -0.0525, -1.8111, -0.8256, -1.1813]])

12个坐标 4个Batch

一个图3个框

∴ 4个图 12个框 ==》 12个坐标