Models for Security Barrier and Face Detection samples not compatible (any longer)?

[Joachim_P_Intel]

Hi,

Getting started with OpenVINO, I installed the latest System Studio and OpenVINO releases (2019.3.376) on an Ubuntu 16.04 box. User local install, i.e. not in /opt/intel. Everything installs, and the samples Security Barrier and Face Detection build with 0 errors and 0 warnings. Next, I used the downloader script to download the corresponding models:

python3 /home/user/intel/openvino/deployment_tools/model_downloader.py --name vehicle-license-plate-detection-barrier-0106, vehicle-attributes-recognition-barrier-0039,license-plate-recognition-barrier-001,face-detection-adas-0001 -o /tmp/OpenVINO

I updated the run configuration to point to input files and model references and export OPEN_VINO_DIR to the local installation directory. Now Run in System Studio

I log into localhost and the programs launch but error out as follows when loading the models (Face Detection for reference, Barrier Sample shows the same problem):

InferenceEngine: 
	API version ............ 2.1
	Build .................. custom_releases/2019/R3_ac8584cb714a697a12f1f30b7a3b78a5b9ac5e05
	Description ....... API
[ INFO ] Parsing input parameters
[ INFO ] Reading input
[ INFO ] Loading plugin CPU

	API version ............ 2.1
	Build .................. 32974
	Description ....... MKLDNNPlugin
[ INFO ] Loading network files for Face Detection
[ INFO ] Batch size is set to 1
[ INFO ] Checking Face Detection network inputs
[ ERROR ] Data is empty!
/opt/intel/openvino_2019.1.094/deployment_tools/inference_engine/include/ie_input_info.hpp:91

Note that the .xml model file is found, but it appears at some point not read correctly along with the .bin file. I am pointing to the FP32 files, the FP16 and INT8 edition error out in the same way.

When first checking the Run configurations, I couldn't help but notice that the default file paths to the model is structured differently from what's downloaded by the downloader. I am under the impression the samples and models have drifted apart and are out of alignment and compatibility.

[JesusE_Intel]

Hi Joachim,

The installation instructions for Intel System Studio includes a Docker file from a previous release 2019 R1. You need to update the Download_Link to point to the latest release (2019.3.376) and update the INSTALLERDIR with the version number.

Please let me know if you have additional questions. 

Regards,

Jesus

[Joachim_P_Intel]

Hi Jesus,

Thank you, that did the trick, the models load now!

Next problem: While the face detection model sample program works now - face detection rectangles are shown in the OpenCV output of the input source (e.g. png image, camera), the security barrier sample does not show any detections. I dialed the confidence level down to 0.15, but no luck.

[JesusE_Intel]

Hi Joachim,

Is the security camera barrier demo showing the image without any detection or not showing the image at all? Are you running the demo from the Intel System Studio? 

Regards,

Jesus

[Joachim_P_Intel]

Yes, images and videos show in the cv::imshow() window, but vehicles (i.e. cars) are not detected and marked with a rectangle or otherwise.

UPDATE. I was a bit quick to jump directly to my own car pictures and videos. The demo-provided car_1.bmp works, i.e. the demo and the models are actually working as advertised. It looks like the car detection is fairly narrow for detecting cars, compared to Mobilenet's, for sure.