@alalek @dkurt I am not sure why it is failing here. these changes are working perfectly fine locally. Any idea how to tackle this?

std::round is C++ feature: https://en.cppreference.com/w/cpp/numeric/math/round
OpenCV 3.4 should use C++98 by default.

Failed builders have GCC 4.x.

@alalek thanks.

@alalek , is there any documentation where I can get to know about the configuration of builders like what version of gcc or C++ custom builder use ?
changes are only getting failed for custom builders.

version of gcc

Check "CMake" step => "stdio" log.

failed for custom builders

These builders are configured to build OpenCV with OpenVINO's Inference engine (Open Source version is here: https://github.com/opencv/dldt/releases/tag/2020.1).

General information about IE and how to build OpenCV with IE locally: https://github.com/opencv/opencv/wiki/Intel%27s-Deep-Learning-Inference-Engine-backend#build-opencv-from-source
This should work well if you using Ubuntu 16.04 / 18.04.

supportBackend

You should tune implementation details through .supportBackend() method. Simplest way is to disable support for both DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019 and DNN_BACKEND_INFERENCE_ENGINE_NGRAPH backend. These implementations can be added later (if possible).

General information about CI options: https://github.com/opencv/opencv/wiki/CI-configuration

@alalek Thank you so much.

Please see the comment #16749 (comment). I think that adding so atomic operations is not a right way to enable object detection networks. They won't cover all the operations from the models and we need more fundamental solution. Can you specify origin frameworks for YOLOv3 and Faster R-CNN models in ONNX format?

@dkurt , all of object detection models are originated from keras2onnx. They are written in keras.

I think that adding so atomic operations is not a right way to enable object detection networks.

It can be useful, for example, for debugging purposes of graph transformations.

@dkurt

we need more fundamental solution.

I have added explanation in #16749 comment . Please share your view.

@alalek , necessary lines of code is already added to PR description to allow custom build then why it is not getting build. Am I missing something ?

@alalek , sorry opencv/opencv_extra#723 had merge conflict yesteday, I have resolved them. can you please let build on custom builders.

I re-triggered all builds, include custom one.

I re-triggered all builds, include custom one.

Thank you so much @asmorkalov . They all are build successfully.

resolves #16749

Please specify if this PR enables one of the models mentioned in the issue

@dkurt Actually I created the issue requesting layer implementation, not model. But as you said that we must focus on models not operators or layers, making models work needs several fusion and layer implementation. It would take time. So if you want me to make model work, please let the PR open. I will complete that. Otherwise this PR can be merged for now since layer implementation is already there and I will work on models separately.

@alalek , I am adding a layer which needs changes in usr/local/include/opencv2/dnn/all_layers.hpp and using those changes more changes are required in modules/dnn/src/layers/pooling_layer.cpp but builder can only find changes in pooling_layer.cpp not in all_layers.hpp.

Am i missing something ?

I am sorry to ask this question here.

usr/local/include

out of source tree.

out of source tree.

ohh.. need to add changes here /modules/dnn/include/opencv2/dnn/all_layers.hpp
Thankyou so much.

I think these changes are not relevant anymore. Closing it for now. I would definitely work on it if find any issue regarding this in future.