tiny-cnn开源库的使用(MNIST)

Posted 2020-06-08 fengbingchun

tiny-cnn是一个基于CNN的开源库，它的License是BSD 3-Clause。作者也一直在维护更新，对进一步掌握CNN很有帮助，因此下面介绍下tiny-cnn在windows7 64bit vs2013的编译及使用。

1.      从https://github.com/nyanp/tiny-cnn下载源码：

$ git clone https://github.com/nyanp/tiny-cnn.git  版本号为77d80a8，更新日期2016.01.22

2.      源文件中已经包含了vs2013工程，vc/tiny-cnn.sln，默认是win32的，examples/main.cpp需要OpenCV的支持，这里新建一个x64的控制台工程tiny-cnn；

3.      仿照源工程，将相应.h文件加入到新控制台工程中，新加一个test_tiny-cnn.cpp文件；

4.      将examples/mnist中test.cpp和train.cpp文件中的代码复制到test_tiny-cnn.cpp文件中；

#include <iostream>
#include <string>
#include <vector>
#include <algorithm>
#include <tiny_cnn/tiny_cnn.h>
#include <opencv2/opencv.hpp>

using namespace tiny_cnn;
using namespace tiny_cnn::activation;

// rescale output to 0-100
template <typename Activation>
double rescale(double x)
{
	Activation a;
	return 100.0 * (x - a.scale().first) / (a.scale().second - a.scale().first);
}

void construct_net(network<mse, adagrad>& nn);
void train_lenet(std::string data_dir_path);
// convert tiny_cnn::image to cv::Mat and resize
cv::Mat image2mat(image<>& img);
void convert_image(const std::string& imagefilename, double minv, double maxv, int w, int h, vec_t& data);
void recognize(const std::string& dictionary, const std::string& filename, int target);

int main()
{
	//train
	std::string data_path = "D:/Download/MNIST";
	train_lenet(data_path);

	//test
	std::string model_path = "D:/Download/MNIST/LeNet-weights";
	std::string image_path = "D:/Download/MNIST/";
	int target[10] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };

	for (int i = 0; i < 10; i++) {
		char ch[15];
		sprintf(ch, "%d", i);
		std::string str;
		str = std::string(ch);
		str += ".png";
		str = image_path + str;

		recognize(model_path, str, target[i]);
	}

	std::cout << "ok!" << std::endl;
	return 0;
}

void train_lenet(std::string data_dir_path) {
	// specify loss-function and learning strategy
	network<mse, adagrad> nn;

	construct_net(nn);

	std::cout << "load models..." << std::endl;

	// load MNIST dataset
	std::vector<label_t> train_labels, test_labels;
	std::vector<vec_t> train_images, test_images;

	parse_mnist_labels(data_dir_path + "/train-labels.idx1-ubyte",
		&train_labels);
	parse_mnist_images(data_dir_path + "/train-images.idx3-ubyte",
		&train_images, -1.0, 1.0, 2, 2);
	parse_mnist_labels(data_dir_path + "/t10k-labels.idx1-ubyte",
		&test_labels);
	parse_mnist_images(data_dir_path + "/t10k-images.idx3-ubyte",
		&test_images, -1.0, 1.0, 2, 2);

	std::cout << "start training" << std::endl;

	progress_display disp(train_images.size());
	timer t;
	int minibatch_size = 10;
	int num_epochs = 30;

	nn.optimizer().alpha *= std::sqrt(minibatch_size);

	// create callback
	auto on_enumerate_epoch = [&](){
		std::cout << t.elapsed() << "s elapsed." << std::endl;
		tiny_cnn::result res = nn.test(test_images, test_labels);
		std::cout << res.num_success << "/" << res.num_total << std::endl;

		disp.restart(train_images.size());
		t.restart();
	};

	auto on_enumerate_minibatch = [&](){
		disp += minibatch_size;
	};

	// training
	nn.train(train_images, train_labels, minibatch_size, num_epochs,
		on_enumerate_minibatch, on_enumerate_epoch);

	std::cout << "end training." << std::endl;

	// test and show results
	nn.test(test_images, test_labels).print_detail(std::cout);

	// save networks
	std::ofstream ofs("D:/Download/MNIST/LeNet-weights");
	ofs << nn;
}

void construct_net(network<mse, adagrad>& nn) {
	// connection table [Y.Lecun, 1998 Table.1]
#define O true
#define X false
	static const bool tbl[] = {
		O, X, X, X, O, O, O, X, X, O, O, O, O, X, O, O,
		O, O, X, X, X, O, O, O, X, X, O, O, O, O, X, O,
		O, O, O, X, X, X, O, O, O, X, X, O, X, O, O, O,
		X, O, O, O, X, X, O, O, O, O, X, X, O, X, O, O,
		X, X, O, O, O, X, X, O, O, O, O, X, O, O, X, O,
		X, X, X, O, O, O, X, X, O, O, O, O, X, O, O, O
	};
#undef O
#undef X

	// construct nets
	nn << convolutional_layer<tan_h>(32, 32, 5, 1, 6)  // C1, [email protected], [email protected]
		<< average_pooling_layer<tan_h>(28, 28, 6, 2)   // S2, [email protected], [email protected]
		<< convolutional_layer<tan_h>(14, 14, 5, 6, 16,
		connection_table(tbl, 6, 16))              // C3, [email protected], [email protected]
		<< average_pooling_layer<tan_h>(10, 10, 16, 2)  // S4, [email protected], [email protected]
		<< convolutional_layer<tan_h>(5, 5, 5, 16, 120) // C5, [email protected], [email protected]
		<< fully_connected_layer<tan_h>(120, 10);       // F6, 120-in, 10-out
}

void recognize(const std::string& dictionary, const std::string& filename, int target) {
	network<mse, adagrad> nn;

	construct_net(nn);

	// load nets
	std::ifstream ifs(dictionary.c_str());
	ifs >> nn;

	// convert imagefile to vec_t
	vec_t data;
	convert_image(filename, -1.0, 1.0, 32, 32, data);

	// recognize
	auto res = nn.predict(data);
	std::vector<std::pair<double, int> > scores;

	// sort & print top-3
	for (int i = 0; i < 10; i++)
		scores.emplace_back(rescale<tan_h>(res[i]), i);

	std::sort(scores.begin(), scores.end(), std::greater<std::pair<double, int>>());

	for (int i = 0; i < 3; i++)
		std::cout << scores[i].second << "," << scores[i].first << std::endl;

	std::cout << "the actual digit is: " << scores[0].second << ", correct digit is: "<<target<<std::endl;

	// visualize outputs of each layer
	//for (size_t i = 0; i < nn.depth(); i++) {
	//	auto out_img = nn[i]->output_to_image();
	//	cv::imshow("layer:" + std::to_string(i), image2mat(out_img));
	//}
	//// visualize filter shape of first convolutional layer
	//auto weight = nn.at<convolutional_layer<tan_h>>(0).weight_to_image();
	//cv::imshow("weights:", image2mat(weight));

	//cv::waitKey(0);
}

// convert tiny_cnn::image to cv::Mat and resize
cv::Mat image2mat(image<>& img) {
	cv::Mat ori(img.height(), img.width(), CV_8U, &img.at(0, 0));
	cv::Mat resized;
	cv::resize(ori, resized, cv::Size(), 3, 3, cv::INTER_AREA);
	return resized;
}

void convert_image(const std::string& imagefilename,
	double minv,
	double maxv,
	int w,
	int h,
	vec_t& data) {
	auto img = cv::imread(imagefilename, cv::IMREAD_GRAYSCALE);
	if (img.data == nullptr) return; // cannot open, or it‘s not an image

	cv::Mat_<uint8_t> resized;
	cv::resize(img, resized, cv::Size(w, h));

	// mnist dataset is "white on black", so negate required
	std::transform(resized.begin(), resized.end(), std::back_inserter(data),
		[=](uint8_t c) { return (255 - c) * (maxv - minv) / 255.0 + minv; });
}

5.      编译时会提示几个错误，解决方法是：

(1)、error C4996，解决方法：将宏_SCL_SECURE_NO_WARNINGS添加到属性的预处理器定义中；

(2)、调用for_函数时，error C2668，对重载函数的调用不明教，解决方法：将for_中的第三个参数强制转化为size_t类型；

6.      运行程序，train时，运行结果如下图所示：

7.      对生成的model进行测试，通过画图工具，每个数字生成一张图像，共10幅，如下图：

通过导入train时生成的model，对这10张图像进行识别，识别结果如下图，其中6和9被误识为5和1：