!1001 fix issue Conv2dBackpropInput bprop should return 3 instead of 2 items

Merge pull request !1001 from zhaozhenlong/conv2d-bp-input-grad-output-3

mindspore/nn/layer/image.py

@@ -58,7 +58,8 @@ class ImageGradients(Cell):
         super(ImageGradients, self).__init__()
 
     def construct(self, images):
-        _check_input_4d(F.shape(images), "images", self.cls_name)
+        check = _check_input_4d(F.shape(images), "images", self.cls_name)
+        images = F.depend(images, check)
         batch_size, depth, height, width = P.Shape()(images)
         dy = images[:, :, 1:, :] - images[:, :, :height - 1, :]
         dy_last = P.Fill()(P.DType()(images), (batch_size, depth, 1, width), 0)

mindspore/ops/_grad/grad_nn_ops.py

@@ -627,7 +627,7 @@ def get_bprop_conv2d_backprop_input(self):
     def bprop(x, w, f_sizes, out, dout):
         dx = input_grad(dout, w)
         dw = filter_grad(x, dout, F.shape(w))
-        return dx, dw
+        return dx, dw, zeros_like(f_sizes)
 
     return bprop