基于预训练Caffe模型的图像分类项目报错求助（VS+OpenCV3.4.0+C++）

Hey there! I see you're working on a Caffe-based image classification project with OpenCV 3.4.0 in Visual Studio, and you're stuck on two common DNN API errors. Let's fix them one by one:

1. "readNet: Identifier not found" #

The readNet function you’re trying to use doesn’t exist in OpenCV 3.4.0 — it was introduced in OpenCV 4.x as a unified function for loading different model formats. In 3.4.0, you need to use the format-specific function for Caffe models.

Solution: #

• Replace readNet with cv::dnn::readNetFromCaffe (this is the correct function for loading Caffe prototxt and model files in 3.4.0)
• Make sure you’ve included the required header:

  #include <opencv2/dnn.hpp>
  

• Alternatively, add using namespace cv::dnn; at the top of your file to avoid typing cv::dnn:: every time (just be cautious with namespace pollution as your project grows).

Example correct call:

cv::dnn::Net net = cv::dnn::readNetFromCaffe("deploy.prototxt", "model.caffemodel");

2. "blobFromImage: function does not take 7 arguments" #

OpenCV 3.4.0’s blobFromImage has a different parameter signature compared to newer OpenCV versions. The 3.4.0 version of the function looks like this:

cv::Mat blobFromImage(
    cv::InputArray image,
    double scalefactor = 1.0,
    const cv::Size& size = cv::Size(),
    const cv::Scalar& mean = cv::Scalar(),
    bool swapRB = false,
    bool crop = false,
    int ddepth = CV_32F
);

The code you copied is likely written for OpenCV 4.x, which added extra parameters (like resizeKeepAspectRatio) or changed parameter order.

Solution: #

Adjust your blobFromImage call to match the 3.4.0 parameter list. For a typical Caffe model (like AlexNet/VGG), a correct preprocessing call would be:

cv::Mat img = cv::imread("input.jpg");
cv::Mat blob = cv::dnn::blobFromImage(
    img,
    1.0,                  // Scale factor
    cv::Size(224, 224),   // Target size matching your Caffe model
    cv::Scalar(104, 117, 123),  // Mean values for BGR (Caffe uses BGR order)
    true,                 // Swap RGB to BGR (OpenCV reads images as BGR)
    false                 // Don't crop the image
);

Quick Additional Checks #

• Double-check your Visual Studio project configuration:
  • Ensure Additional Include Directories includes <your-opencv-path>/include and <your-opencv-path>/include/opencv2
  • Ensure Additional Library Directories includes <your-opencv-path>/lib
  • Add the correct DNN library to Linker > Input: use opencv_dnn340.lib for release mode, opencv_dnn340d.lib for debug mode (or opencv_world340.lib/opencv_world340d.lib if using the world package)

内容的提问来源于stack exchange，提问作者Sanjay Kumar