I am trying to run MTCNN Face detection on intel NCS2 using OpenVino toolkit. I am having the IR files for PNet, RNet and ONet which are used for MTCNN Face detection. I am able to load the .xml and .bin files on MYRIAD Device.
When an image is provided to the network using infer() function, an array is received as the output. I am unable to manipulate that array to proceed further.
We apologize for the inconvenience, we already reported a bug for the web page. Regarding your original issue, we are currently looking into it, we will come back to you when we get an update on this. Could you please answer the following:
- Which OpenVINO™ toolkit version are you using?
- Which OS are you using?
- Could you provide us with your IR files and source code and the commands necessary to test it from our end?
- If there is any kind of error, please share it.
The details you require are as follows:
OpenVino toolkit version: 2020.2.120
OS: Ubuntu 18.04.4
I have attached the IR files of mtcnn-pnet and the source code (mtcnn_face_detection.py) which I am using to fetch output from mtcnn-pnet model.
I am providing a face image to the p-net model and also receiving an output, but I am not sure whether this output is correct or not and I also don't know how to use the output further to act as an input to the r-net and o-net model.
I also tried to convert the p-net, r-net and o-net (caffe model) files to a single tensorflow model, but I am unable to convert it further to IR format using Model Optimizer. I have also attached this tensorflow model i.e mtcnn.pb
Thanks and Regards,
Thank you for the information given.
The output of the mtcnn-p model is a set of softmax probabilities as shown here. This comment on an earlier post related to your question tells you what to do with the outputs of each model. Additionally, you can check this article that tells you more about the outputs between each layer.