(5.x) Merge 4.x

testdata/cv/stereomatching/algorithms/cuda_stereosgm_params.xml

@@ -0,0 +1,22 @@
+<?xml version="1.0"?>
+<opencv_storage>
+<params>
+	<!-- caseName, datasetName, numDisp, mode -->
+	case_barn2_1 barn2 "64" "1"
+	case_barn2_3 barn2 "64" "3"
+	case_bull_1 bull "64" "1"
+	case_bull_3 bull "64" "3"
+	case_cones_1 cones "64" "1"
+	case_cones_3 cones "64" "3"
+	case_poster_1 poster "64" "1"
+	case_poster_3 poster "64" "3"
+	case_sawtooth_1 sawtooth "64" "1"
+	case_sawtooth_3 sawtooth "64" "3"
+	case_teddy_1 teddy "64" "1"
+	case_teddy_3 teddy "64" "3"
+	case_tsukuba_1 tsukuba "64" "1"
+	case_tsukuba_3 tsukuba "64" "3"
+	case_venus_1 venus "64" "1"
+	case_venus_3 venus "64" "3"
+</params>
+</opencv_storage>

testdata/dnn/darknet/generate_darknet_models.py

@@ -36,4 +36,5 @@ def genTestData(name, inpShape, outShape):
 genTestData('route', [1, 4, 3, 6], [1, 2, 2, 3])
 genTestData('route_multi', [1, 6, 3, 6], [1, 4, 2, 3])
 genTestData('relu', [1, 2, 3, 4], [1, 2, 3, 4])
+genTestData('sam', [1, 2, 3, 4], [1, 4, 3, 4])
 

testdata/dnn/darknet/sam.cfg

@@ -0,0 +1,22 @@
+[net]
+batch=1
+channels=2
+height=3
+width=4
+
+[convolutional]
+filters=4
+size=3
+stride=1
+pad=1
+activation=leaky
+
+[convolutional]
+filters=4
+size=1
+stride=1
+pad=1
+activation=logistic
+
+[sam]
+from=-2

testdata/dnn/darknet/tanh.cfg

@@ -0,0 +1,12 @@
+[net]
+batch=1
+channels=2
+height=2
+width=2
+
+# Just an identity layer to enable relative reference of shortcut layer.
+[maxpool]
+size=1
+stride=1
+activation=tanh
+

testdata/dnn/download_models.py

@@ -61,7 +61,7 @@ def get(self):
             return True
 
         basedir = os.path.dirname(self.filename)
-        if not os.path.exists(basedir):
+        if basedir and not os.path.exists(basedir):
             print('  creating directory: ' + basedir)
             os.makedirs(basedir, exist_ok=True)
 
@@ -864,6 +864,16 @@ def get_confirm_token(response):  # in case of large files
         downloader=GDrive('1j4UTqVE4EGaUFiK7a5I_CYX7twO9c5br'),
         sha='49776d262993c387542f84d9cd16566840404f26',
         filename='gsoc2016-goturn/goturn.caffemodel'),
+    Model(
+        name='crnn',
+        url='https://drive.google.com/uc?export=dowload&id=1ooaLR-rkTl8jdpGy1DoQs0-X0lQsB6Fj',
+        sha='270d92c9ccb670ada2459a25977e8deeaf8380d3',
+        filename='onnx/models/crnn.onnx'),
+    Model(
+        name='DB_TD500_resnet50',
+        url='https://drive.google.com/uc?export=dowload&id=19YWhArrNccaoSza0CfkXlA8im4-lAGsR',
+        sha='1b4dd21a6baa5e3523156776970895bd3db6960a',
+        filename='onnx/models/DB_TD500_resnet50.onnx'),
 ]
 
 # Note: models will be downloaded to current working directory

testdata/dnn/onnx/generate_onnx_models.py

@@ -378,17 +378,62 @@ def forward(self, x):
 
 class Slice(nn.Module):
 
-    def __init__(self):
+    def __init__(self, custom_slice=None):
+        self.custom_slice=custom_slice
         super(Slice, self).__init__()
 
     def forward(self, x):
+        if self.custom_slice:
+           return x[self.custom_slice]
+
         return x[..., 1:-1, 0:3]
 
 input = Variable(torch.randn(1, 2, 4, 4))
 model = Slice()
 save_data_and_model("slice", input, model)
 save_data_and_model("slice_opset_11", input, model, version=11)
 
+input_2 = Variable(torch.randn(6, 6))
+custom_slice_list = [
+    slice(1, 3, 1),
+    slice(0, 3, 2)
+]
+model_2 = Slice(custom_slice=custom_slice_list)
+save_data_and_model("slice_opset_11_steps_2d", input_2, model_2, version=11)
+postprocess_model("models/slice_opset_11_steps_2d.onnx", [['height', 'width']])
+
+input_3 = Variable(torch.randn(3, 6, 6))
+custom_slice_list_3 = [
+    slice(None, None, 2),
+    slice(None, None, 2),
+    slice(None, None, 2)
+]
+model_3 = Slice(custom_slice=custom_slice_list_3)
+save_data_and_model("slice_opset_11_steps_3d", input_3, model_3, version=11)
+postprocess_model("models/slice_opset_11_steps_3d.onnx", [[3, 'height', 'width']])
+
+input_4 = Variable(torch.randn(1, 3, 6, 6))
+custom_slice_list_4 = [
+    slice(0, 5, None),
+    slice(None, None, None),
+    slice(1, None, 2),
+    slice(None, None, None)
+]
+model_4 = Slice(custom_slice=custom_slice_list_4)
+save_data_and_model("slice_opset_11_steps_4d", input_4, model_4, version=11)
+postprocess_model("models/slice_opset_11_steps_4d.onnx", [["batch_size", 3, 'height', 'width']])
+
+input_5 = Variable(torch.randn(1, 2, 3, 6, 6))
+custom_slice_list_5 = [
+    slice(None, None, None),
+    slice(None, None, None),
+    slice(0, None, 3),
+    slice(None, None, None),
+    slice(None, None, 2)
+]
+model_5 = Slice(custom_slice=custom_slice_list_5)
+save_data_and_model("slice_opset_11_steps_5d", input_5, model_5, version=11)
+
 class Eltwise(nn.Module):
 
     def __init__(self):
@@ -958,6 +1003,14 @@ def forward(self, x):
 model = Power(2)
 save_data_and_model("pow2", x, model)
 
+class Exp(nn.Module):
+  def forward(self, x):
+    return x.exp()
+
+x = Variable(torch.randn(2, 2))
+model = Exp()
+save_data_and_model("exp", x, model)
+
 class ReduceMaxGlobal(nn.Module):
   def forward(self, x):
     out = torch.max(x)
@@ -1224,4 +1277,37 @@ def forward(self, x):
 
 x = Variable(torch.randn(1, 2, 20, 20))
 model = Conv1ResizePoold()
-save_data_and_model("conv_resize_pool_1d", x, model)
+save_data_and_model("conv_resize_pool_1d", x, model)
+
+class Mish(nn.Module):
+    def forward(self, x):
+        return x * (torch.tanh(F.softplus(x)))
+
+x = Variable(torch.randn([1, 2, 2, 2]))
+model = Mish()
+save_data_and_model("mish", x, model)
+
+class PadCalculation(nn.Module):
+    def forward(self, x):
+        y = F.max_pool2d(x, kernel_size=2)
+        diff_h = x.shape[2] - y.shape[2]
+        diff_w = x.shape[3] - y.shape[3]
+        y = F.pad(y, [diff_w // 2, diff_w - diff_w // 2, diff_h // 2, diff_h - diff_h // 2])
+        return y
+
+x = Variable(torch.randn([1, 1, 3, 4]))
+model = PadCalculation()
+save_data_and_model("calc_pads", x, model, version=11)
+
+class NormalizeFusion(nn.Module):
+    def forward(self, x):
+        mul = x * x
+        sum = torch.sum(mul, dim=(1), keepdim=True)
+        maximum = torch.clamp(sum, min=1e-8)
+        sqrt = torch.sqrt(maximum)
+        reciprocal = torch.reciprocal(sqrt)
+        return x * reciprocal
+
+x = Variable(torch.randn([2, 3]))
+model = NormalizeFusion()
+save_data_and_model("normalize_fusion", x, model)

testdata/dnn/onnx/models/exp.onnx

@@ -0,0 +1,11 @@
+pytorch1.7:P
+
+
0
1Exp_0"Exptorch-jit-exportZ
+
0
+

+
+b
+
1
+

+
+B	

testdata/dnn/onnx/models/mish.onnx

@@ -0,0 +1,21 @@
+pytorch1.7:�

+
+
0
1
+Softplus_0"Softplus
+
+
1
2Tanh_1"Tanh
+
+
0
+
2
3Mul_2"Multorch-jit-exportZ
+
0
+

+

+
+
+b
+
3
+

+

+
+
+B