[Fix] cast the type of mask to enable training with amp (#2220)

mmcv/ops/modulated_deform_conv.py

@@ -68,6 +68,7 @@ class ModulatedDeformConv2dFunction(Function):
         input = input.type_as(offset)
         weight = weight.type_as(input)
         bias = bias.type_as(input)  # type: ignore
+        mask = mask.type_as(input)
         ctx.save_for_backward(input, offset, mask, weight, bias)
         output = input.new_empty(
             ModulatedDeformConv2dFunction._output_size(ctx, input, weight))