Accuracy of OpenCV DeepFlow V.S. FlowNet 2.0
FlowNet 2.0 seems to be widely used and regarded as the state of the art (?) in the community. I am wondering if anyone can provide any insights on its accuracy comparing to DeepFlow in OpenCV. Setting up a working python environment or making the pre-trained flownet 2.0 model work with OpenCV's DNN module is not so straight forward for me. I am wondering if it's worth the effort to figure them out.
Just try running the model at all. Sometimes the model has custom layers(written as custom (python)functions) which could break things.
And yes its worth trying - you can use opencv as top lvl api for various models. And now that we have dnn cuda support in the latest master...