[Fix] Add Pytorch HardSwish assertion in unit test (open-mmlab#1294)

* assert original HardSwish when PyTorch > 1.6 in unit test

* assert original HardSwish when PyTorch > 1.6 in unit test

* assert original HardSwish when PyTorch > 1.6 in unit test

* assert original HardSwish when PyTorch > 1.6 in unit test

* assert original HardSwish when PyTorch > 1.6 in unit test

* assert original HardSwish when PyTorch > 1.6 in unit test

Author: MengzhangLI

tests/test_models/test_backbones/test_blocks.py

@@ -2,6 +2,7 @@
 import mmcv
 import pytest
 import torch
+from mmcv.utils import TORCH_VERSION, digit_version
 
 from mmseg.models.utils import (InvertedResidual, InvertedResidualV3, SELayer,
                                 make_divisible)
@@ -108,19 +109,34 @@ def test_inv_residualv3():
     assert inv_module.expand_conv.conv.kernel_size == (1, 1)
     assert inv_module.expand_conv.conv.stride == (1, 1)
     assert inv_module.expand_conv.conv.padding == (0, 0)
-    assert isinstance(inv_module.expand_conv.activate, mmcv.cnn.HSwish)
 
     assert isinstance(inv_module.depthwise_conv.conv,
                       mmcv.cnn.bricks.Conv2dAdaptivePadding)
     assert inv_module.depthwise_conv.conv.kernel_size == (3, 3)
     assert inv_module.depthwise_conv.conv.stride == (2, 2)
     assert inv_module.depthwise_conv.conv.padding == (0, 0)
     assert isinstance(inv_module.depthwise_conv.bn, torch.nn.BatchNorm2d)
-    assert isinstance(inv_module.depthwise_conv.activate, mmcv.cnn.HSwish)
+
     assert inv_module.linear_conv.conv.kernel_size == (1, 1)
     assert inv_module.linear_conv.conv.stride == (1, 1)
     assert inv_module.linear_conv.conv.padding == (0, 0)
     assert isinstance(inv_module.linear_conv.bn, torch.nn.BatchNorm2d)
+
+    if (TORCH_VERSION == 'parrots'
+            or digit_version(TORCH_VERSION) < digit_version('1.7')):
+        # Note: Use PyTorch official HSwish
+        # when torch>=1.7 after MMCV >= 1.4.5.
+        # Hardswish is not supported when PyTorch version < 1.6.
+        # And Hardswish in PyTorch 1.6 does not support inplace.
+        # More details could be found from:
+        # https://github.com/open-mmlab/mmcv/pull/1709
+        assert isinstance(inv_module.expand_conv.activate, mmcv.cnn.HSwish)
+        assert isinstance(inv_module.depthwise_conv.activate, mmcv.cnn.HSwish)
+    else:
+        assert isinstance(inv_module.expand_conv.activate, torch.nn.Hardswish)
+        assert isinstance(inv_module.depthwise_conv.activate,
+                          torch.nn.Hardswish)
+
     x = torch.rand(1, 32, 64, 64)
     output = inv_module(x)
     assert output.shape == (1, 40, 32, 32)