BlobFromImages for gpuMat

C++
,
1

Hi! How to apply BlobFromImages for gpuMat?

2

There is no specific function for CUDA. Just set Target and backend

    net = dnn::readNet("hed_deploy.prototxt", "hed_pretrained_bsds.caffemodel");
    net_cuda.setPreferableTarget(DNN_TARGET_CUDA);
    net_cuda.setPreferableBackend(DNN_BACKEND_CUDA);
    net_cuda.setInput(inp);

Then You can make a benchmark
Tps cuda :12.2785ms
Tps opencv :360.703ms
Tps inference :95.4307ms
Tps vino :232.38ms

for my configuration

3

My model has different format. BlobFromImages is just function for preprocessing

4

I don’t understand.
You make what you want with your data and use setInp.
But to infer with GPU it is setPreferableBackend() and setPreferableTarget()

5

I ment that I don’t use dnn for net parsing and inference. In my pipeline I need to preprocess gpu-frame with BlobFromImages (or find some similar function) and then feed it to tensorrt-inference.

6

tensorrt-inference? That’s not opencv function
Is your question about opencv? → google is your friend