Implementation:Tencent Ncnn YOLOv2 Example

Knowledge Sources	Tencent_Ncnn
Domains	Vision, Object Detection
Last Updated	2026-02-09 19:00 GMT

Overview

Concrete tool for object detection on PASCAL VOC classes using YOLOv2 with a MobileNet-YOLO backbone and ncnn.

Description

This example demonstrates YOLOv2 object detection using a MobileNet backbone (MobileNet-YOLO) with the ncnn framework. It loads the pretrained model from the MobileNet-YOLO project, resizes input images to 416x416, and applies Caffe-YOLOv2-style normalization: first scaling by 0.007843 (1/127.5) and then subtracting a mean of 1.0. This two-step preprocessing is distinct from other SSD examples. The model extracts the "detection_out" layer and parses results in the standard (label, prob, x1, y1, x2, y2) format for the 20 PASCAL VOC classes.

Usage

Use this example for YOLO-based object detection when a lightweight MobileNet backbone is preferred over the original Darknet architecture. It is an early YOLO-series example suitable for understanding the YOLOv2 detection pipeline with ncnn and for comparing against the SSD-based detection examples.

Code Reference

Source Location

• Repository: Tencent_Ncnn
• File: examples/yolov2.cpp
• Lines: 1-147

Signature

static int detect_yolov2(const cv::Mat& bgr, std::vector<Object>& objects);
static void draw_objects(const cv::Mat& bgr, const std::vector<Object>& objects);
int main(int argc, char** argv);

Import

#include "net.h"

I/O Contract

Inputs

Name	Type	Required	Description
image_path	const char*	Yes	Path to input image (passed as command-line argument)
bgr	const cv::Mat&	Yes	BGR image loaded by OpenCV, resized internally to 416x416

Outputs

Name	Type	Description
objects	std::vector<Object>	Detected objects with label, prob, and rect for VOC classes
Visual output	cv::imshow window	Displayed image with bounding boxes and class labels

Model Files

File	Description
mobilenet_yolo.param	MobileNet-YOLO network parameter file
mobilenet_yolo.bin	MobileNet-YOLO network weight file

Preprocessing

• Resize: Input resized to 416x416 using ncnn::Mat::from_pixels_resize with PIXEL_BGR
• Two-step normalization (Caffe-YOLOv2 style):
  • Step 1: Scale by norm values [0.007843, 0.007843, 0.007843] (i.e., 1/127.5)
  • Step 2: Subtract mean values [1.0, 1.0, 1.0]
• This is equivalent to: X' = X * 0.007843 - 1.0

Usage Examples

Running the Example

./yolov2 image.jpg

Key Code Pattern

ncnn::Net yolov2;
yolov2.opt.use_vulkan_compute = true;

yolov2.load_param("mobilenet_yolo.param");
yolov2.load_model("mobilenet_yolo.bin");

ncnn::Mat in = ncnn::Mat::from_pixels_resize(bgr.data, ncnn::Mat::PIXEL_BGR,
    bgr.cols, bgr.rows, 416, 416);

// Caffe-YOLOv2-Windows style: X' = X * scale - mean
const float mean_vals[3] = {1.0f, 1.0f, 1.0f};
const float norm_vals[3] = {0.007843f, 0.007843f, 0.007843f};
in.substract_mean_normalize(0, norm_vals);
in.substract_mean_normalize(mean_vals, 0);

ncnn::Extractor ex = yolov2.create_extractor();
ex.input("data", in);

ncnn::Mat out;
ex.extract("detection_out", out);

Related Pages

• Environment:Tencent_Ncnn_Build_Environment