Question about multi_shifts_pts_tensor

[Watebear]

May I ask a question about variable: multi_shifts_pts_tensor. It's shape is [num_samples, num_final_shift_num, fixed_num, 2], in practice [n, 19, 20, 2]。 In dimension 1， 19 is consists of 3 parts：（1）1 is coordinate points of line，（2）1 is coordinate points of line in reverse order，（3）other 17 are padding values with -1。 Why is it designed like this？

[LinaShanghaitech]

May I ask a question about variable: multi_shifts_pts_tensor. It's shape is [num_samples, num_final_shift_num, fixed_num, 2], in practice [n, 19, 20, 2]。 In dimension 1， 19 is consists of 3 parts：（1）1 is coordinate points of line，（2）1 is coordinate points of line in reverse order，（3）other 17 are padding values with -1。 Why is it designed like this？

I think the author wants to align the different numbers of permutations between Polyline and Polygon, which you can check Figure.3 in the paper. Polyline only has two permutations, so it is needed to pad with an invalid value (-10000).

[Watebear]

t

Thanks！

[LegendBC]

It seems that the issue has been resolved and thanks for the helpful discussion.