[object detection][mtcnn]

[kao__mars]

Hi~

The CV SDK have any plan to support the mtcnn model with tensorflow version (only for caffe version now)?

And have any sample code for using ir mtcnn model (convert from caffe mtcnn) on CV SDK? 

Thanks.

[JongIl_P_Intel]

Currently OpenVINO is supporting:

Caffe - MTCNN-PNet MTCNN-RNet MTCNN-ONet

Inference Engine supports  MKLDNN(CPU), clDNN(GEN) for MTCNN.

Thanks,

Jeff

[lemniscate]

In most implementations of MTCNN, the image is passed at multiple scales to the p-net.

However,

request = net.CreateInferRequestPtr();
Blob::Ptr  blog = request->GetBlob(input);
SizeVector blobSize = blob->getTensorDesc().getDims();
const size_t width = blobSize[3];
const size_t height = blobSize[2];
const size_t channels = blobSize[1];
T* blob_data = blob->buffer().as<T*>();
cv::Mat resized_image(orig_image);
if (width != orig_image.size().width || height!= orig_image.size().height) {
        cv::resize(orig_image, resized_image, cv::Size(width, height));
}

will resize the input image to the size of the input layer.

How to pass images at different sizes to p-net? p-net is fully convolutional, can the input size not be fixed?