mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-29 05:29:54 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Import Upsample and Unsqueeze from ONNX

Browse Source
This commit is contained in:
Dmitry Kurtaev 2019-02-21 19:48:46 +03:00
parent 1db5d82b7f
commit 20400aa9f7
2 changed files with 62 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
modules/dnn/src/onnx/onnx_importer.cpp
View File

@ -392,10 +392,10 @@ void ONNXImporter::populateNet(Net dstNet)
layerParams.set("ceil_mode", isCeilMode(layerParams));
layerParams.set("ave_pool_padded_area", framework_name == "pytorch");
}
else if (layer_type == "GlobalAveragePool")
else if (layer_type == "GlobalAveragePool" || layer_type == "GlobalMaxPool")
{
layerParams.type = "Pooling";
layerParams.set("pool", "AVE");
layerParams.set("pool", layer_type == "GlobalAveragePool" ? "AVE" : "MAX");
layerParams.set("global_pooling", true);
}
else if (layer_type == "Add" || layer_type == "Sum")
@ -448,6 +448,11 @@ void ONNXImporter::populateNet(Net dstNet)
layerParams.set("bias_term", false);
}
}
else if (layer_type == "Neg")
{
layerParams.type = "Power";
layerParams.set("scale", -1);
}
else if (layer_type == "Constant")
{
CV_Assert(node_proto.input_size() == 0);
@ -595,9 +600,12 @@ void ONNXImporter::populateNet(Net dstNet)
else if (layer_type == "Unsqueeze")
{
CV_Assert(node_proto.input_size() == 1);
DictValue axes = layerParams.get("axes");
if (constBlobs.find(node_proto.input(0)) != constBlobs.end())
{
// Constant input.
Mat input = getBlob(node_proto, constBlobs, 0);
DictValue axes = layerParams.get("axes");
std::vector<int> dims;
for (int j = 0; j < input.dims; j++) {
dims.push_back(input.size[j]);
@ -611,6 +619,17 @@ void ONNXImporter::populateNet(Net dstNet)
constBlobs.insert(std::make_pair(layerParams.name, out));
continue;
}
// Variable input.
if (axes.size() != 1)
CV_Error(Error::StsNotImplemented, "Multidimensional unsqueeze");
int dims[] = {1, -1};
layerParams.type = "Reshape";
layerParams.set("axis", axes.getIntValue(0));
layerParams.set("num_axes", 1);
layerParams.set("dim", DictValue::arrayInt(&dims[0], 2));
}
else if (layer_type == "Reshape")
{
CV_Assert(node_proto.input_size() == 2 || layerParams.has("shape"));
@ -707,6 +726,25 @@ void ONNXImporter::populateNet(Net dstNet)
continue;
}
}
else if (layer_type == "Upsample")
{
layerParams.type = "Resize";
if (layerParams.has("scales"))
{
// Pytorch layer
DictValue scales = layerParams.get("scales");
CV_Assert(scales.size() == 4);
layerParams.set("zoom_factor_y", scales.getIntValue(2));
layerParams.set("zoom_factor_x", scales.getIntValue(3));
}
else
{
// Caffe2 layer
replaceLayerParam(layerParams, "height_scale", "zoom_factor_y");
replaceLayerParam(layerParams, "width_scale", "zoom_factor_x");
}
replaceLayerParam(layerParams, "mode", "interpolation");
}
else
{
for (int j = 0; j < node_proto.input_size(); j++) {

10
modules/dnn/test/test_onnx_importer.cpp
View File

@ -140,6 +140,11 @@ TEST_P(Test_ONNX_layers, Padding)
testONNXModels("padding");
}
TEST_P(Test_ONNX_layers, Resize)
{
testONNXModels("resize_nearest");
}
TEST_P(Test_ONNX_layers, MultyInputs)
{
const String model = _tf("models/multy_inputs.onnx");
@ -169,6 +174,11 @@ TEST_P(Test_ONNX_layers, DynamicReshape)
testONNXModels("dynamic_reshape");
}
TEST_P(Test_ONNX_layers, Reshape)
{
testONNXModels("unsqueeze");
}
INSTANTIATE_TEST_CASE_P(/*nothing*/, Test_ONNX_layers, dnnBackendsAndTargets());
class Test_ONNX_nets : public Test_ONNX_layers {};