cv::VideoCapture with flag cv::CAP_INTEL_MFX fails to open

[SURABHI_G_Intel]

Hello,

I have openvino_2019.1.148 installed on my windows machine. I am working on interactive_face_detection_demo sample. Currently the sample can offload only face detection or other deep learning models supported by sample on GPU. But the capture of frame from camera with openCV still executes on CPU. In order to offload even this to GPU so that frame directly gets captured into memory accessible by integrated gfx I was trying to use Intel media SDK support. But the call cap.open(0, cv::CAP_INTEL_MFX) fails. I am new to openVINO as well as Intel media SDK. Could you please let me know why this call is failing or what would be the right way to offload capture from camera part on integrated GPU.

-Surabhi

[Shubha_R_Intel]

Dear SURABHI G. ,

First I hope I can convince you to upgrade to 2019R2 which just got released yesterday.

Secondly, this forum is for Model Optimizer and Inference Engine questions. Please post OpenCV questions (your question is clearly an OpenCV category question) to here:

https://answers.opencv.org/questions/

https://github.com/opencv/opencv/issues

Thanks,

Shubha

[SURABHI_G_Intel]

Thanks Shubha. Sure I will upgrade. My question is more about integration of Intel Media SDK with openvino toolkit. Could you please point me to right forum for that.

In my installation of openVINO i don't see Intel media SDK installed by default as suggested by OpenVINO documentation.

[SURABHI_G_Intel]

I just checked announcements section for this forum. "Use this forum to ask questions and share information with others about the OpenVINO™ toolkit (formerly Intel® Computer Vision SDK), OpenCV* and all things computer vision-related on Intel® platforms."

So I think my question fits here. Hope I can get some help with the issue I am facing.

[Maksim_S_Intel · Accepted solution]

There is no MediaSDK videoio support in OpenCV in OpenVINO 2019R1. Please try the latest release 2019R2 it has experimental MediaSDK plugin for videoio module.

Also, MediaSDK support in OpenCV is very basic, it does not support media containers (e.g. avi, mkv) and always copies frame content to host memory. So in order to achieve full GPU-based processing, you have to use MediaSDK directly, I believe there should be sample applications.