Backward compat fix for Pytorch < 0.4 (#472)

Author: fmassa

torchvision/utils.py

@@ -58,7 +58,7 @@ def norm_range(t, range):
             if range is not None:
                 norm_ip(t, range[0], range[1])
             else:
-                norm_ip(t, t.min().item(), t.max().item())
+                norm_ip(t, float(t.min()), float(t.max()))
 
         if scale_each is True:
             for t in tensor:  # loop over mini-batch dimension