Queries about failed call “u2net_human_seg.onnx” in c++ about opencv_zoo

[jsxyhelu]

There is Python code that can be run for "u2net_human_seg. Onnx"

   # Inference
    net.setInput(blob)
    d0 = net.forward()
    # Norm
    pred = normPred(d0[:, 0, :, :])

There are problems in rewriting to C + +. My code is like this

    Mat inputBlob = blobFromImages(src, 1.0/255.0, Size(320, 320),Scalar(), true, false);
    net.setInput(inputBlob);
    Mat prob = net.forward();

The error echo is:

“[ INFO:0] global E:\Gitee\opencv_2021\opencv\modules\dnn\src\onnx\onnx_importer.cpp (395) cv::dnn::dnn4_v20201117::ONNXImporter::populateNet DNN/ONNX: loading ONNX v6 model produced by 'pytorch':1.8. Number of nodes = 1055, inputs = 1, outputs = 7
OpenCV(4.5.1-dev) Error: Assertion failed (total(os[i]) > 0) in cv::dnn::dnn4_v20201117::Net::Impl::getLayerShapesRecursively, file E:\Gitee\opencv_2021\opencv\modules\dnn\src\dnn.cpp, line 3520”

I tried to do some research, but I can't find the reason. I hope I can help with C + + calling. Thanks!

[fengyuentau]

Can you provide the related cpp files so that I can help debugging?

[jsxyhelu]

OK，cpp code is simple

#include <iostream>
#include "opencv2/dnn.hpp"
#include "opencv2/imgproc.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/objdetect.hpp"
using namespace cv;
using namespace std;
using namespace cv::dnn;

int main(int argc, char ** argv)
{
	Net net = readNetFromONNX("e:/template/u2net_human_seg.onnx");
	Mat src = imread("e:/template/lena.jpg");
	if (net.empty()) {
		printf("read  model data failure...\n");
		return -1;
	}
	Mat inputBlob = blobFromImages(src, 1.0/255.0, Size(320, 320),Scalar(), true, false);
	net.setInput(inputBlob);
	Mat prob = net.forward();
	//ERROR
	/*
	[ INFO:0] global E:\Gitee\opencv_2021\opencv\modules\dnn\src\onnx\onnx_importer.cpp (395) cv::dnn::dnn4_v20201117::ONNXImporter::populateNet DNN/ONNX: loading ONNX v6 model produced by 'pytorch':1.8. Number of nodes = 1055, inputs = 1, outputs = 7
OpenCV(4.5.1-dev) Error: Assertion failed (total(os[i]) > 0) in cv::dnn::dnn4_v20201117::Net::Impl::getLayerShapesRecursively, file E:\Gitee\opencv_2021\opencv\modules\dnn\src\dnn.cpp, line 3520”
	*/
	return 0;
}

I think this is the same as this Python code

    net = cv.dnn.readNet(args.model)
    input_size = 320 # fixed
    # build blob using OpenCV
    img = cv.imread(args.input)
    blob = cv.dnn.blobFromImage(img, scalefactor=(1.0/255.0), size=(input_size, input_size), swapRB=True)
    # Inference
    net.setInput(blob)
    d0 = net.forward()

[ztxc]

hello, Is your problem solved？