Removed float indices from dnn tests

testdata/dnn/onnx/generate_onnx_models.py

@@ -2665,7 +2665,7 @@ def __init__(self, *args, **kwargs):
         super(ArgMax, self).__init__()
 
     def forward(self, x):
-        return torch.argmax(x, dim=2, keepdims=False).to(torch.float32)
+        return torch.argmax(x, dim=2, keepdims=False)
 
 model = ArgMax()
 input_ = Variable(torch.randn(2, 3, 4, 5, dtype=torch.float32))
@@ -2676,7 +2676,7 @@ def __init__(self, *args, **kwargs):
         super(ArgMin, self).__init__()
 
     def forward(self, x):
-        return torch.argmin(x, dim=-1, keepdims=True).to(torch.float32)
+        return torch.argmin(x, dim=-1, keepdims=True)
 
 model = ArgMin()
 input_ = Variable(torch.randn(2, 3, 4, 5, dtype=torch.float32))

testdata/dnn/tensorflow/generate_tf2_models.py

@@ -28,7 +28,7 @@ def writeBlob(data, name, nchw = False):
         # NDHWC->NCDHW
         data = data.transpose(0, 4, 1, 2, 3)
 
-    data = np.ascontiguousarray(data.astype(np.float32))
+    data = np.ascontiguousarray(data)
     np.save(name + '.npy', data)
 
 
@@ -148,6 +148,18 @@ def saveBroken(graph, name):
 y = tf.compat.v1.nn.conv2d_backprop_input(input_sizes=tf.constant([1, 3, 4, 2]), filter=kernel, out_backprop=x, data_format = "NHWC", padding = [[0, 0], [2, 1], [2, 1], [0, 0]], strides = [1, 3, 2, 1])
 model = tf.keras.Model(x, y)
 save(model, 'conv2d_backprop_input_asymmetric_pads_nhwc', False, x=tf.TensorSpec(shape=(1, 2, 3, 3), dtype=tf.float32))
+################################################################################
+tf.keras.backend.set_image_data_format('channels_first')
+x = tf.keras.layers.Input(batch_shape = (2, 3, 4), name='x')
+y = tf.math.argmax(x, axis=-1)
+model = tf.keras.Model(x, y)
+save(model, 'argmax', True, x=tf.TensorSpec(shape=(2, 3, 4), dtype=tf.float32))
+################################################################################
+tf.keras.backend.set_image_data_format('channels_last')
+x = tf.keras.layers.Input(batch_shape = (2, 3, 4), name='x')
+y = tf.math.argmin(x, axis=1)
+model = tf.keras.Model(x, y)
+save(model, 'argmin', False, x=tf.TensorSpec(shape=(2, 3, 4), dtype=tf.float32))
 
 # Uncomment to print the final graph.
 # with tf.io.gfile.GFile('tf2_prelu_net.pb', 'rb') as f:

testdata/dnn/tensorflow/generate_tf_models.py

@@ -1070,14 +1070,6 @@ def pad_depth(x, desired_channels):
 square = tf.square(inp)
 save(inp, square, 'square')
 ################################################################################
-inp = tf.placeholder(tf.float32, [2, 3, 4], 'input')
-argmax = tf.argmax(inp, -1)
-save(inp, argmax, 'argmax')
-################################################################################
-inp = tf.placeholder(tf.float32, [2, 3, 4], 'input')
-argmin = tf.argmin(inp, 1)
-save(inp, argmin, 'argmin')
-################################################################################
 # Generate graph and test data for Reshape permutations check
 stride = 1
 kernel_size = 3