Merge remote-tracking branch 'upstream/3.4' into merge-3.4

Author: alalek

testdata/dnn/onnx/data/input_ave_pool3d.npy

@@ -263,3 +263,23 @@ def forward(self, x):
 model = Unsqueeze()
 model.eval()
 save_data_and_model("unsqueeze", input, model)
+
+input = Variable(torch.randn(1, 2, 4, 5))
+deconv_adjpad2d = nn.ConvTranspose2d(2, 3, (3, 2), stride=(1, 2), padding=(1, 2), output_padding=(0, 1))
+save_data_and_model("deconv_adjpad_2d", input, deconv_adjpad2d)
+
+input = Variable(torch.randn(1, 2, 3, 4, 5))
+conv3d = nn.Conv3d(2, 3, (2, 3, 2), stride=(1, 1, 1), padding=(0, 0, 0), groups=1, dilation=(1, 1, 1), bias=False)
+save_data_and_model("conv3d", input, conv3d)
+
+input = Variable(torch.randn(1, 2, 3, 4, 5))
+conv3d = nn.Conv3d(2, 3, (2, 3, 3), stride=(1, 2, 3), padding=(0, 1, 2), groups=1, dilation=(1, 2, 3), bias=True)
+save_data_and_model("conv3d_bias", input, conv3d)
+
+input = torch.randn(1, 2, 3, 4, 6)
+maxpool3d = nn.MaxPool3d((3, 2, 5), stride=(2, 1, 2), padding=(1, 0, 2))
+save_data_and_model("max_pool3d", input, maxpool3d)
+
+input = torch.randn(1, 2, 3, 5, 6)
+avepool3d = nn.AvgPool3d((3, 4, 3), stride=(1, 2, 3), padding=(1, 2, 0))
+save_data_and_model("ave_pool3d", input, avepool3d)

testdata/dnn/onnx/data/input_conv3d.npy

@@ -44,6 +44,9 @@ def writeBlob(data, name):
     if data.ndim == 4:
         # NHWC->NCHW
         np.save(name + '.npy', data.transpose(0, 3, 1, 2).astype(np.float32))
+    elif data.ndim == 5:
+        # NDHWC->NCDHW
+        np.save(name + '.npy', data.transpose(0, 4, 1, 2, 3).astype(np.float32))
     else:
         # Save raw data.
         np.save(name + '.npy', data.astype(np.float32))
@@ -724,6 +727,44 @@ def pad_depth(x, desired_channels):
 
     with tf.gfile.FastGFile('slim_softmax_v2_net.pb', 'wb') as f:
         f.write(graph_def.SerializeToString())
+################################################################################
+inp = tf.placeholder(tf.float32, [1, 4, 6, 5, 3], 'input') # NDHWC format
+conv3d = tf.layers.conv3d(inputs=inp, filters=2, kernel_size=[2, 3, 4], use_bias=True, padding='same')
+save(inp, conv3d, 'conv3d')
+################################################################################
+inp = tf.placeholder(tf.float32, [1, 4, 6, 5, 3], 'input') # NDHWC format
+maxpool3d = tf.layers.max_pooling3d(inputs=inp, pool_size=(3, 2, 3), strides=(1, 2, 1), padding='same')
+save(inp, maxpool3d, 'max_pool3d')
+################################################################################
+inp = tf.placeholder(tf.float32, [1, 5, 4, 5, 2], 'input') # NDHWC format
+avepool3d = tf.layers.average_pooling3d(inputs=inp, pool_size=(3, 3, 2), strides=(2, 1, 1), padding='valid')
+save(inp, avepool3d, 'ave_pool3d')
+################################################################################
+# issue https://github.com/opencv/opencv/issues/13494
+inp_node = 'input_image'
+out_node = 'SUBPIXEL/SUBPIXEL/subpixel_image/Identity'
+with tf.Session(graph=tf.Graph()) as localSession:
+    localSession.graph.as_default()
+
+    with tf.gfile.FastGFile('simple_subpixel.optimized.pb') as f:
+        graph_def = tf.GraphDef()
+        graph_def.ParseFromString(f.read())
+
+    tf.import_graph_def(graph_def, name='')
+
+    inputData = gen_data(tf.placeholder(tf.float32, [1, 1, 1, 4], inp_node))
+    outputData = localSession.run(localSession.graph.get_tensor_by_name(out_node + ':0'),
+                                  feed_dict={inp_node + ':0': inputData})
+    writeBlob(inputData, 'subpixel_in')
+    writeBlob(outputData, 'subpixel_out')
+
+    for node in graph_def.node:
+        if node.op == 'Placeholder':
+            node.attr["data_format"].s = "NHWC"
+
+    with tf.gfile.FastGFile('subpixel_net.pb', 'wb') as f:
+        f.write(graph_def.SerializeToString())
+################################################################################
 
 # Uncomment to print the final graph.
 # with tf.gfile.FastGFile('fused_batch_norm_net.pb') as f: