System information (version)

• OpenCV => 4.2.0 (via OpenCVSharp https://github.com/shimat/opencvsharp realesed on 08.01.2020)
• Operating System / Platform => Windows 64 Bit
• Compiler => Visual Studio 2017 (C#)

Detailed description

I have tried to use FlowNet 2.0 (https://github.com/lmb-freiburg/flownet2) to calculate the Optical Flow between two images. I'm setting two inputs and changed the Resolutions in the prototxt.template to the resolution i want to use. Every time i call Forward i got the Exception "inputs.size() >= 2".

Steps to reproduce

• Downloaded the model "flownet2-models.tar.gz" contained in the shell script https://github.com/lmb-freiburg/flownet2/blob/master/models/download-models.sh
• Using the largest model contained in the folder "FlowNet2".
• Changed the Placeholder Variables (Resolutions) in the prototext to the one i want to use. Based on the logic in https://github.com/lmb-freiburg/flownet2/blob/master/scripts/run-flownet.py
  • My variant of the file: FlowNet2_deploy.prototxt.txt
• Run Forward like the C#-Code below.

left and right are some images read from an video-file or just two images loaded with ImRead. I think i should not attach some files during copyright problems.

using (var net = CvDnn.ReadNetFromCaffe(@"FlowNet2_deploy.prototxt.txt",
    @"FlowNet2_weights.caffemodel.h5"))
{
    net.SetPreferableBackend(Net.Backend.OPENCV);
    net.SetPreferableTarget(Net.Target.CPU);

    var img0 = CvDnn.BlobFromImage(left, 1.0, new Size(1024, 720), new Scalar(), true, false);
    var img1 = CvDnn.BlobFromImage(right, 1.0, new Size(1024, 720), new Scalar(), true, false);
    net.SetInput(img0, "img0");
    net.SetInput(img1, "img1");

    var predict = net.Forward("predict_flow_final");  << exception
}

I have used an face-detection model based on Caffe with one input and this worked fine.