Some confusion about v10postprocess.

[IronMan1111]

Why is topk used twice in v10postprocess? In my opinion, if we modify v10postprocess to perform a single topk calculation as shown below, the result would be the same.

def v10postprocess(preds, max_det, nc=80):
    assert(4 + nc == preds.shape[-1])
    boxes, scores = preds.split([4, nc], dim=-1)
    scores, index = torch.topk(scores.flatten(1), max_det, dim=-1)
    labels = index % nc
    index = index // nc
    boxes = boxes.gather(dim=1, index=index.unsqueeze(-1).repeat(1, 1, boxes.shape[-1]))
    return boxes, scores, labels