Added few tests for torch

Author: arrybn

modules/dnn/samples/torch_enet.cpp

@@ -23,6 +23,7 @@ const String keys =
         "{c_names c || path to file with classnames for channels (optional, categories.txt) }"
         "{result r  || path to save output blob (optional, binary format, NCHW order) }"
         "{show s    || whether to show all output channels or not}"
+        "{o_blob    || output blob's name. If empty, last blob's name in net is used}"
         ;
 
 std::vector<String> readClassNames(const char *filename);
@@ -112,7 +113,13 @@ int main(int argc, char **argv)
 
     //! [Gather output]
 
-    dnn::Blob prob = net.getBlob(net.getLayerNames().back());   //gather output of "prob" layer
+    String oBlob = net.getLayerNames().back();
+    if (!parser.get<String>("o_blob").empty())
+    {
+        oBlob = parser.get<String>("o_blob");
+    }
+
+    dnn::Blob prob = net.getBlob(oBlob);   //gather output of "prob" layer
 
     Mat& result = prob.matRef();
 

modules/dnn/src/layers/max_unpooling_layer.cpp

@@ -28,6 +28,8 @@ void MaxUnpoolLayerImpl::allocate(const std::vector<Blob*> &inputs, std::vector<
     outShape[2] = outSize.height;
     outShape[3] = outSize.width;
 
+    CV_Assert(inputs[0]->total() == inputs[1]->total());
+
     outputs.resize(1);
     outputs[0].create(outShape);
 }

modules/dnn/test/test_torch_importer.cpp

@@ -72,7 +72,8 @@ TEST(Torch_Importer, simple_read)
     importer->populateNet(net);
 }
 
-static void runTorchNet(String prefix, String outLayerName, bool isBinary)
+static void runTorchNet(String prefix, String outLayerName = "",
+                        bool check2ndBlob = false, bool isBinary = false)
 {
     String suffix = (isBinary) ? ".dat" : ".txt";
 
@@ -92,52 +93,69 @@ static void runTorchNet(String prefix, String outLayerName, bool isBinary)
     Blob out = net.getBlob(outLayerName);
 
     normAssert(outRef, out);
+
+    if (check2ndBlob)
+    {
+        Blob out2 = net.getBlob(outLayerName + ".1");
+        Blob ref2 = readTorchBlob(_tf(prefix + "_output_2" + suffix), isBinary);
+        normAssert(out2, ref2);
+    }
 }
 
 TEST(Torch_Importer, run_convolution)
 {
-    runTorchNet("net_conv", "l1_Convolution", false);
+    runTorchNet("net_conv");
 }
 
 TEST(Torch_Importer, run_pool_max)
 {
-    runTorchNet("net_pool_max", "l1_Pooling", false);
+    runTorchNet("net_pool_max", "", true);
 }
 
 TEST(Torch_Importer, run_pool_ave)
 {
-    runTorchNet("net_pool_ave", "l1_Pooling", false);
+    runTorchNet("net_pool_ave");
 }
 
 TEST(Torch_Importer, run_reshape)
 {
-    runTorchNet("net_reshape", "l1_Reshape", false);
-    runTorchNet("net_reshape_batch", "l1_Reshape", false);
+    runTorchNet("net_reshape");
+    runTorchNet("net_reshape_batch");
 }
 
 TEST(Torch_Importer, run_linear)
 {
-    runTorchNet("net_linear_2d", "l1_InnerProduct", false);
+    runTorchNet("net_linear_2d");
 }
 
 TEST(Torch_Importer, run_paralel)
 {
-    runTorchNet("net_parallel", "l2_torchMerge", false);
+    runTorchNet("net_parallel", "l2_torchMerge");
 }
 
 TEST(Torch_Importer, run_concat)
 {
-    runTorchNet("net_concat", "l2_torchMerge", false);
+    runTorchNet("net_concat", "l2_torchMerge");
 }
 
 TEST(Torch_Importer, run_deconv)
 {
-    runTorchNet("net_deconv", "", false);
+    runTorchNet("net_deconv");
 }
 
 TEST(Torch_Importer, run_batch_norm)
 {
-    runTorchNet("net_batch_norm", "", false);
+    runTorchNet("net_batch_norm");
+}
+
+TEST(Torch_Importer, net_prelu)
+{
+    runTorchNet("net_prelu");
+}
+
+TEST(Torch_Importer, net_cadd_table)
+{
+    runTorchNet("net_cadd_table");
 }
 
 #if defined(ENABLE_TORCH_ENET_TESTS)

modules/dnn/testdata/dnn/torch/torch_gen_test_data.lua

@@ -27,6 +27,8 @@ function save(net, input, label)
 	torch.save(label .. '_input.txt', input, 'ascii')
 	--torch.save(label .. '_output.dat', output)
 	torch.save(label .. '_output.txt', output, 'ascii')
+
+	return net
 end
 
 local net_simple = nn.Sequential()
@@ -38,7 +40,8 @@ save(net_simple, torch.Tensor(2, 3, 25, 35), 'net_simple')
 
 local net_pool_max = nn.Sequential()
 net_pool_max:add(nn.SpatialMaxPooling(4,5, 3,2, 1,2):ceil()) --TODO: add ceil and floor modes
-save(net_pool_max, torch.rand(2, 3, 50, 30), 'net_pool_max')
+local net = save(net_pool_max, torch.rand(2, 3, 50, 30), 'net_pool_max')
+torch.save('net_pool_max_output_2.txt', net.modules[1].indices - 1, 'ascii')
 
 local net_pool_ave = nn.Sequential()
 net_pool_ave:add(nn.SpatialAveragePooling(4,5, 2,1, 1,2))
@@ -74,5 +77,15 @@ net_deconv:add(nn.SpatialFullConvolution(3, 9, 4, 5, 1, 2, 0, 1, 0, 1))
 save(net_deconv, torch.rand(2, 3, 4, 3) - 0.5, 'net_deconv')
 
 local net_batch_norm = nn.Sequential()
-net_batch_norm:add(nn.SpatialBatchNormalization(3))
-save(net_batch_norm, torch.rand(1, 3, 4, 3) - 0.5, 'net_batch_norm')
+net_batch_norm:add(nn.SpatialBatchNormalization(4, 1e-3))
+save(net_batch_norm, torch.rand(1, 4, 5, 6) - 0.5, 'net_batch_norm')
+
+local net_prelu = nn.Sequential()
+net_prelu:add(nn.PReLU(5))
+save(net_prelu, torch.rand(1, 5, 40, 50) - 0.5, 'net_prelu')
+
+local net_cadd_table = nn.Sequential()
+local sum = nn.ConcatTable()
+sum:add(nn.Identity()):add(nn.Identity())
+net_cadd_table:add(sum):add(nn.CAddTable())
+save(net_cadd_table, torch.rand(1, 5, 40, 50) - 0.5, 'net_cadd_table')