[Relay][Expr] how can I swap inputs' data's column in relay?


How can I swap inputs’ data’s column in relay?
I try to use non_max_suppression op in onnx frontend.
The relay’s non_max_suppression op’s first input requires a 3-d tensor with shape [batch, num, 6], but in onnx, nms’ input boxes shape is [batch, spatial_dim, 4] and score’s shape is [batch, num_class, spatial_dim], so I try to concatenate boxes and score into [batch, num, 6].
Is there any hint about this?