[Relay][Frontend][TF] fix _parse_param bug (apache#4711)

python/tvm/relay/frontend/tensorflow.py

@@ -2391,7 +2391,7 @@ def _parse_param(self, key, value, name, shape):
             if np_array.dtype == np.dtype(object):
                 # Object types are generally tensorflow DT_STRING (DecodeJpeg op).
                 # Just leave it as placeholder.
-                if shape:
+                if shape and name in shape:
                     var_shape = shape[name]
                 else:
                     var_shape = tensor_util.TensorShapeProtoToList(value.tensor.tensor_shape)

tests/python/frontend/tensorflow/test_debugging.py

@@ -20,19 +20,22 @@
 from tvm import relay
 from tvm.relay.frontend.tensorflow import from_tensorflow
 
-def run_relay(graph, *vars):
-    mod, params = from_tensorflow(graph.as_graph_def(add_shapes=True))
+def run_relay(graph, shape_dict=None, *vars):
+    mod, params = from_tensorflow(
+        graph.as_graph_def(add_shapes=True),
+        shape=shape_dict)
     ex = relay.create_executor('debug', mod=mod)
     return ex.evaluate()(*vars)
 
 def test_assert_true():
     g = tf.Graph()
+    shape = (1, 2)
     with g.as_default():
-        x = tf.placeholder(tf.float32, shape=())
-        assert_op = tf.Assert(tf.less_equal(x, x), ["it failed"])
+        x = tf.placeholder(tf.float32, shape=shape, name="input")
+        assert_op = tf.Assert(tf.reduce_all(tf.less_equal(x, x)), ["it failed"])
 
         with tf.Session() as sess:
-            x_value = np.random.rand()
+            x_value = np.random.rand(*shape)
             assert sess.run(assert_op, feed_dict={x: x_value}) is None
 
         # In TVM, tf.assert is converted to a no-op which is actually a 0,
@@ -44,7 +47,7 @@ def test_assert_true():
         # do that, it's happening in Relay, and that optimization shouldn't
         # affect the arity of the main function. We should have to pass in
         # x_value here.
-        np.testing.assert_allclose(0, run_relay(g).asnumpy())
+        np.testing.assert_allclose(0, run_relay(g, {'input':shape}).asnumpy())
 
 def test_assert_true_var_capture():
     g = tf.Graph()
@@ -65,7 +68,8 @@ def test_assert_true_var_capture():
         # the graph as a boolean, which is not correct - as you can see above,
         # TF believes that the value of this graph is None. In addition, the
         # arity of the translated function should be 1, not 2.
-        np.testing.assert_allclose(True, run_relay(g, x_value, x_value).asnumpy())
+        np.testing.assert_allclose(True,
+            run_relay(g, None, x_value, x_value).asnumpy())
 
 def test_assert_false():
     g = tf.Graph()