[Q] GDL - Regular Group Convolutions trilinear sampling

[sasasqt]

Tutorial: DL2-1 (Fill-in number of tutorial)

Describe the bug
grid = torch.roll(grid, shifts=1, dims=-1) in trilinear sampling

Expected behavior
grid = torch.flip(grid, dims=[-1])

Screenshots
visualization on number 6:
using roll

---

vs using flip