Implementation:Tencent Ncnn MobileNetV2 SSDLite 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 MobileNetV2 SSD-Lite with ncnn.

Description

This example demonstrates MobileNetV2 SSD-Lite object detection using the ncnn framework. It loads a pretrained MobileNetV2-SSDLite model (originally from Caffe), resizes input images to 300x300 with mean/norm preprocessing, and extracts detection results for the 20 PASCAL VOC classes. A custom Noop layer is registered as a stub for the "Silence" layer from the original Caffe model that is not natively supported by ncnn. Detected objects are drawn with bounding boxes and class labels using OpenCV.

Usage

Use this example when you need to perform multi-class object detection on images using a lightweight MobileNetV2-based SSD-Lite model. It is suitable for mobile and edge deployment scenarios where detection of common VOC categories (person, car, dog, etc.) is required. It also serves as a reference for registering custom layer stubs when converting models with unsupported layers.

Code Reference

Source Location

• Repository: Tencent_Ncnn
• File: examples/mobilenetv2ssdlite.cpp
• Lines: 1-150

Signature

static int detect_mobilenetv2(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 300x300

Outputs

Name	Type	Description
objects	std::vector<Object>	Detected objects, each with label (int), prob (float), and rect (cv::Rect_<float>)
Visual output	cv::imshow window	Displayed image with bounding boxes and class labels

Model Files

File	Description
mobilenetv2_ssdlite_voc.param	Network parameter file
mobilenetv2_ssdlite_voc.bin	Network weight file

Preprocessing

• Resize: Input resized to 300x300 using ncnn::Mat::from_pixels_resize with PIXEL_BGR
• Mean values: [127.5, 127.5, 127.5]
• Norm values: [1/127.5, 1/127.5, 1/127.5]
• Custom layer: Registers "Silence" layer mapped to a Noop stub via DEFINE_LAYER_CREATOR(Noop)

Usage Examples

Running the Example

./mobilenetv2ssdlite image.jpg

Key Code Pattern

ncnn::Net mobilenetv2;
mobilenetv2.opt.use_vulkan_compute = true;
mobilenetv2.register_custom_layer("Silence", Noop_layer_creator);

mobilenetv2.load_param("mobilenetv2_ssdlite_voc.param");
mobilenetv2.load_model("mobilenetv2_ssdlite_voc.bin");

ncnn::Mat in = ncnn::Mat::from_pixels_resize(bgr.data, ncnn::Mat::PIXEL_BGR,
    bgr.cols, bgr.rows, 300, 300);

const float mean_vals[3] = {127.5f, 127.5f, 127.5f};
const float norm_vals[3] = {1.0 / 127.5, 1.0 / 127.5, 1.0 / 127.5};
in.substract_mean_normalize(mean_vals, norm_vals);

ncnn::Extractor ex = mobilenetv2.create_extractor();
ex.input("data", in);

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

Related Pages

• Environment:Tencent_Ncnn_Build_Environment