Thanks!

Can this method test the devices? I mean which lists it will show in the following scenarios:

• When there is a device but there is no plugin available (i.e. CPU with renamed MKLDNNPlugin)
• When there is a plugin but there is no device (i.e. MYRIAD)
• When there is a device and plugin but it's unavailable (i.e. with MYRIAD which is used by different process)
• When there is no plugin and there is no device (i.e. ARM CPU as in case of Raspberry Pi).

Thanks!

Can this method test the devices? I mean which lists it will show in the following scenarios:

• When there is a device but there is no plugin available (i.e. CPU with renamed MKLDNNPlugin)

• this is handled by OpenVino: no plugin means it won't show up in the list of available devices

• When there is a plugin but there is no device (i.e. MYRIAD)

• this is handled by OpenVino: no device means it won't show up in the list of available devices

• When there is a device and plugin but it's unavailable (i.e. with MYRIAD which is used by different process)

• I believe this is not handled but I'm unable to reproduce this right now

• When there is no plugin and there is no device (i.e. ARM CPU as in case of Raspberry Pi).

• handled as well, no plugin -> not in the list

Can this or similar method determine Myriad 2 or Myriad X?

Can you also experiment replacing checkIETarget from dnn.cpp using this new function? Thanks!

Well I did not know I could call cv::dnn::getAvailableBackends() to do this
In op_inf_engine.cpp there's a isMyriadX() function

@Nefast, May I ask you if getAvailableTargets(DNN_BACKEND_INFERENCE_ENGINE) can match your expectations? This way we could add a wrapper to it instead.

If yes, you can modify modify this PR in the following way: (it'll also require to reopen PR with 3.4 branch as a target, we will port the changes to master branch after that).

diff --git a/modules/dnn/include/opencv2/dnn/dnn.hpp b/modules/dnn/include/opencv2/dnn/dnn.hpp
index 94e2ada..c5d3975 100644
--- a/modules/dnn/include/opencv2/dnn/dnn.hpp
+++ b/modules/dnn/include/opencv2/dnn/dnn.hpp
@@ -94,7 +94,7 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
      */
     enum Target
     {
-        DNN_TARGET_CPU,
+        DNN_TARGET_CPU = 0,
         DNN_TARGET_OPENCL,
         DNN_TARGET_OPENCL_FP16,
         DNN_TARGET_MYRIAD,
@@ -102,7 +102,7 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
     };
 
     CV_EXPORTS std::vector< std::pair<Backend, Target> > getAvailableBackends();
-    CV_EXPORTS std::vector<Target> getAvailableTargets(Backend be);
+    CV_EXPORTS_W std::vector<Target> getAvailableTargets(Backend be);
 
     /** @brief This class provides all data needed to initialize layer.
      *
diff --git a/modules/dnn/misc/python/pyopencv_dnn.hpp b/modules/dnn/misc/python/pyopencv_dnn.hpp
index 34aeacb..69c1424 100644
--- a/modules/dnn/misc/python/pyopencv_dnn.hpp
+++ b/modules/dnn/misc/python/pyopencv_dnn.hpp
@@ -71,6 +71,12 @@ PyObject* pyopencv_from(const dnn::LayerParams& lp)
     return dict;
 }
 
+template<>
+PyObject* pyopencv_from(const std::vector<dnn::Target> &t)
+{
+    return pyopencv_from(std::vector<int>(t.begin(), t.end()));
+}
+
 class pycvLayer CV_FINAL : public dnn::Layer
 {
 public:

@dkurt Do you mean I should replace calls to bool checkIETarget(Target target) https://github.com/opencv/opencv/blob/master/modules/dnn/src/dnn.cpp#L110 with my solution?
Now that you showed me that I can use getAvailableTargets(), my PR doesn't make sense anymore

@Nefast, You still can help us adding a wrapper for getAvailableTargets because for now it's C++ only. So you can do it in this PR (or it's even better to reopen it to 3.4 branch with a different branch name):

git checkout 3.4
git checkout -b branch_name

Sorry to be slow. You want me to reopen a PR on 3.4 with nothing more than the diff you posted just above?

@Nefast, Yes, if you don't mind.

Ok. I close this PR as a duplicate of #16232