update FID docstring (#2497)

Co-authored-by: Desroziers <sylvain.desroziers@michelin.com>

Author: sdesrozis

ignite/metrics/gan/fid.py

@@ -67,7 +67,7 @@ class FID(_BaseInceptionMetric):
 
     Remark:
 
-        This implementation is inspired by pytorch_fid package which can be found `here`__
+        This implementation is inspired by `pytorch_fid` package which can be found `here`__
 
         __ https://github.com/mseitzer/pytorch-fid
 
@@ -114,6 +114,47 @@ class FID(_BaseInceptionMetric):
 
             0.0
 
+    .. note::
+
+        The default `torchvision` model used is InceptionV3 pretrained on ImageNet.
+        This can lead to differences in results with `pytorch_fid`. To find comparable results,
+        the following model wrapper should be used:
+
+        .. code::
+
+            import torch.nn as nn
+
+            # wrapper class as feature_extractor
+            class WrapperInceptionV3(nn.Module):
+
+                def __init__(self, fid_incv3):
+                    super().__init__()
+                    self.fid_incv3 = fid_incv3
+
+                @torch.no_grad()
+                def forward(self, x):
+                    y = self.fid_incv3(x)
+                    y = y[0]
+                    y = y[:, :, 0, 0]
+                    return y
+
+            # use cpu rather than cuda to get comparable results
+            device = "cpu"
+
+            # pytorch_fid model
+            dims = 2048
+            block_idx = InceptionV3.BLOCK_INDEX_BY_DIM[dims]
+            model = InceptionV3([block_idx]).to(device)
+
+            # wrapper model to pytorch_fid model
+            wrapper_model = WrapperInceptionV3(model)
+            wrapper_model.eval();
+
+            # comparable metric
+            pytorch_fid_metric = FID(num_features=dims, feature_extractor=wrapper_model)
+
+        Important, `pytorch_fid` results depend on the batch size if the device is `cuda`.
+
     .. versionadded:: 0.4.6
     """