未分配内存的指针被释放异常

[zhuzhu18]

平台(如果交叉编译请再附上交叉编译目标平台):

Platform(Include target platform as well if cross-compiling):

mac os 11.7.10

代码如下：

C++
//
// Created by xxx on 19.9.24.
//

#include<MNN/Interpreter.hpp>
#include<MNN/ImageProcess.hpp>
#include<MNN/expr/Expr.hpp>
#include<MNN/expr/Executor.hpp>
#include<MNN/expr/ExprCreator.hpp>
#include<MNN/expr/Module.hpp>
#include<MNN/AutoTime.hpp>
#include
#include
#include
#include
#include

class SESR{
public:
SESR()=default;
int Init(const char* model_path);
std::vectorMNN::Express::VARP pre_process(const uint8_t * yuv_data, int width, int height, int channels);
std::vectorMNN::Express::VARP& processing(std::vectorMNN::Express::VARP& yuv_var);
uint8_t* post_process(std::vectorMNN::Express::VARP& yuv_var);

private:
const float mean_vals[1] = {0.f};
const float normal_vals[1] = {1.f/255};
std::shared_ptrMNN::CV::ImageProcess pretreat;
std::shared_ptrMNN::Express::Module net = nullptr;
};

int SESR::Init(const char* model_path) {
net.reset(MNN::Express::Module::load(std::vectorstd::string{"img"}, std::vectorstd::string{"out"}, model_path));
if(nullptr == net){
MNN_ERROR("load model failed!\n");
}
MNN::ScheduleConfig sConfig;
// sConfig.type = MNN_FORWARD_CPU;
sConfig.type = MNN_FORWARD_OPENCL;
sConfig.backupType = MNN_FORWARD_VULKAN;
sConfig.numThread = std::thread::hardware_concurrency(); // cpu推理时的线程数
MNN::BackendConfig backendConfig;
backendConfig.memory = MNN::BackendConfig::Memory_Normal;
backendConfig.precision = MNN::BackendConfig::Precision_Normal;
sConfig.backendConfig = &backendConfig;

// std::shared_ptrMNN::Express::Executor::RuntimeManager rtmgr{
// MNN::Express::Executor::RuntimeManager::createRuntimeManager(sConfig)
// };
// if(nullptr == rtmgr){
// MNN_ERROR("Empty RuntimeManager\n");
// }
MNN::CV::ImageProcess::Config config;
config.sourceFormat = MNN::CV::YUV_I420; // 指定源格式YUV_I420
config.destFormat = MNN::CV::YUV_I420; // 转换成的目标格式YUV_I420
std::memcpy(config.mean, mean_vals, sizeof(mean_vals));
std::memcpy(config.normal, normal_vals, sizeof(normal_vals));
pretreat = std::shared_ptrMNN::CV::ImageProcess{MNN::CV::ImageProcess::create(config)};
}

std::vectorMNN::Express::VARP SESR::pre_process(const uint8_t* yuv_data, int width, int height, int channels) {
// MNN::Express::Variable::Info y_info{MNN::Express::NCHW, {1, channels, height, width},
// halide_type_of(), channelsheightwidth};
// MNN::Express::EXPRP y_expr = MNN::Express::Expr::create(std::move(y_info), yuv_data, MNN::Express::VARP::INPUT, MNN::Express::Expr::REF);
// MNN::Express::VARP y = MNN::Express::Variable::create(y_expr);
MNN::Express::VARP y = MNN::Express::_Input({1, channels, height, width}, MNN::Express::NCHW); // 模型输入格式NCHW, RGB
pretreat->convert(yuv_data, width, height, widthchannels, y->writeMap(),
width, height, channels, widthchannels, halide_type_of()); // y通道数据

MNN::Express::Variable::Info u_info{MNN::Express::NCHW, {1, channels, height/2, width/2},
                                    halide_type_of<uint8_t>(), channels*height*width/4};
MNN::Express::EXPRP u_expr = MNN::Express::Expr::create(std::move(u_info), yuv_data+channels*height*width, MNN::Express::VARP::INPUT, MNN::Express::Expr::REF);
MNN::Express::VARP u = MNN::Express::Variable::create(u_expr);       // u通道数据

MNN::Express::Variable::Info v_info{MNN::Express::NCHW, {1, channels, height/2, width/2},
                                    halide_type_of<uint8_t>(), channels*height*width/4};
MNN::Express::EXPRP v_expr = MNN::Express::Expr::create(std::move(v_info), yuv_data+channels*height*width*5/4, MNN::Express::VARP::INPUT, MNN::Express::Expr::REF);
MNN::Express::VARP v = MNN::Express::Variable::create(v_expr);       // v通道数据
y = y * MNN::Express::_Scalar<float>(1.f/255);
std::vector<MNN::Express::VARP> yuv_var = {y, u, v};
return yuv_var;

}

std::vectorMNN::Express::VARP& SESR::processing(std::vectorMNN::Express::VARP& yuv_var) {
yuv_var[0] = net->onForward({yuv_var[0]})[0]; // 取出第一个输出
yuv_var[1] = MNN::Express::_Resize(yuv_var[1], 1.5, 1.5); // 对u通道数据插值缩放
yuv_var[2] = MNN::Express::_Resize(yuv_var[2], 1.5, 1.5); // 对v通道数据插值缩放

return yuv_var;

}

uint8_t* SESR::post_process(std::vectorMNN::Express::VARP& yuv_var) {
yuv_var[0] = MNN::Express::_Relu6(yuv_var[0], 0.0f, 1.0f); // 截断到[0, 255]
yuv_var[0] = yuv_var[0] * MNN::Express::_Scalar(255.0f); // 像素值缩放到[0, 255]之间
yuv_var[0] = MNN::Express::_Cast<uint8_t>(yuv_var[0]); // 像素值转换为整数
yuv_var[0] = MNN::Express::_Reshape(yuv_var[0], {-1});
yuv_var[1] = MNN::Express::_Reshape(yuv_var[1], {-1});
yuv_var[2] = MNN::Express::_Reshape(yuv_var[2], {-1});

MNN::Express::VARP out_var = MNN::Express::_Concat(yuv_var, 0);
auto* out_img = out_var->readMap<uint8_t>();

return const_cast<uint8_t*>(out_img);

}

int main(){
const std::string& file_path = "/Users/xxx/pycharm_projects/mnn_practice/720p_I420.yuv";
const std::string& dest_path = "/Users/xxx/pycharm_projects/mnn_practice/out_720p.yuv";
const std::string& model_path = "/Users/xxx/pycharm_projects/mnn_practice/sesr_c1.mnn";
u_int32_t width = 1280;
u_int32_t height = 720;
u_int32_t channels = 1;
u_int32_t num_pixels = width * height * channels * 1.5; // 每个像素有3个通道（R, G, B）

// 分配内存来存储RGB数据
uint8_t* yuv_data = new uint8_t[num_pixels];

// 打开文件
std::ifstream file(file_path, std::ios::binary);
if (!file.is_open()) {
    std::cerr << "读取文件数据失败: " << file_path << std::endl;
    delete[] yuv_data;
    return 1;
}
// 读取数据到内存
file.read(reinterpret_cast<char*>(yuv_data), num_pixels);
file.close();

std::shared_ptr<SESR> model {new SESR()};
model->Init(model_path.c_str());        // 加载模型文件, 设置cpu/gpu/npu推理, 线程数等

std::vector<MNN::Express::VARP> yuv_var = model->pre_process(yuv_data, width, height, channels);      // 从连续内存的数组中读取yuv三个通道的数据写入相应的VARP中, 并放入一个vector中

yuv_var = model->processing(yuv_var);     // y通道通过模型超分, u和v通道直接插值

uint8_t* out_img = model->post_process(yuv_var);     // 输出截断至[0, 255], 并取整, RGB->BGR
// 打开文件写入
std::ofstream outFile(dest_path, std::ios::binary | std::ios::out);
if (!outFile.is_open()) {
    std::cerr << "无法创建文件: " << dest_path << std::endl;
    return 1;
}
num_pixels = yuv_var[0]->getInfo()->size + yuv_var[1]->getInfo()->size + yuv_var[2]->getInfo()->size;   // yuv三个通道的总像素个数
// 写入 yuv 数据
outFile.write(reinterpret_cast<char*>(out_img), num_pixels);

outFile.close();
delete[] yuv_data;

return 0;

}

报的异常截图如下：

模型文件：https://github.com/zhuzhu18/sesr/blob/main/sesr_c1.mnn

[jxt1234]

可以先用 Xcode 开 asan 测试一下