[Fix] Fix warning with torch.meshgrid. (open-mmlab#860)

* Fix warning with torch.meshgrid

* Add torch_meshgrid_ij wrapper

* Use `digit_version` instead of packaging package.

Co-authored-by: mzr1996 <mzr1996@163.com>

tests/test_models/test_utils/test_attention.py

@@ -1,18 +1,25 @@
 # Copyright (c) OpenMMLab. All rights reserved.
+from functools import partial
 from unittest import TestCase
 from unittest.mock import ANY, MagicMock
 
 import pytest
 import torch
+from mmcv.utils import TORCH_VERSION, digit_version
 
 from mmcls.models.utils.attention import ShiftWindowMSA, WindowMSA
 
+if digit_version(TORCH_VERSION) >= digit_version('1.10.0a0'):
+    torch_meshgrid_ij = partial(torch.meshgrid, indexing='ij')
+else:
+    torch_meshgrid_ij = torch.meshgrid  # Uses indexing='ij' by default
+
 
 def get_relative_position_index(window_size):
     """Method from original code of Swin-Transformer."""
     coords_h = torch.arange(window_size[0])
     coords_w = torch.arange(window_size[1])
-    coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+    coords = torch.stack(torch_meshgrid_ij([coords_h, coords_w]))  # 2, Wh, Ww
     coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
     # 2, Wh*Ww, Wh*Ww
     relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]