创建垫子后，根据字节向量将图像加载到垫子上

Question

My situation is the next: I'm working on a multi-threaded application in which differents threads receive images in the form of bytes from sockets, store them in a vector<char> and should generate a cv::Mat with it in order to be able to do the processing with OpenCV . Well, the complications are mainly because in order to separate the private data of every thread from the one of the others I used an array of structs , which I declared with all the data structures that every thread may need. Thus I also have to declare the vector<char> and the cv::Mat inside of it, and once the vector has received all the image bytes I have to find a way to 'load them' appropriately inside the cv::Mat without creating any additional one , because the memory space reserved to every thread is fixed.

---

What I've tried so far:

struct thread_data {                   //private data for threads
     std::vector<char> buf_img;
     Mat img_scene;
     Mat img_temp;
     //...

     thread_data() { 
          buf_img.reserve(65000),      // initialization with enough space
      img_scene.create(700,500, CV_8U),
      img_temp.create(700,500, CV_8U);
          }
    };


thread_data *tdata;
// declare array of structs
tdata = (thread_data * ) calloc(nthreads, sizeof(thread_data));

//...

// And the thread function, once received the image:

private_tm->img_temp = cv::Mat(private_tm->buf_img,true).clone(); //from vector to Mat

if( !private_tm->img_temp.data ) { 
      std::cout<< " --(!) Image could not be read " << std::endl; 
      private_tm->answer = "error";
}
else {
    // decode image and call the function passing the pointer to the struct:
    private_tm->img_scene = cv::Mat(cv::imdecode(private_tm->temp,1)).clone();
private_tm->answer = OCV_func((void*)private_tm); 
 }

---

Result after receiving one single image:

> Thread 2:
> ==12142== Invalid write of size 4
> ==12142==    at 0x804B869: cv::Mat::release() (mat.hpp:369)
> ==12142==    by 0x804EEDD: cv::Mat::operator=(cv::Mat const&) (mat.hpp:287)
> ==12142==    by 0x804E8A2: thread_main(void*) (threads.cpp:408)
> ==12142==    by 0x45A4D4B: start_thread (pthread_create.c:308)
> ==12142==    by 0x46A7DDD: clone (clone.S:130)
> ==12142==  Address 0x0 is not stack'd, malloc'd or (recently) free'd
> ==12142== 
> ==12142== 
> ==12142== Process terminating with default action of signal 11 (SIGSEGV)
> ==12142==  Access not within mapped region at address 0x0
> ==12142==    at 0x804B869: cv::Mat::release() (mat.hpp:369)
> ==12142==    by 0x804EEDD: cv::Mat::operator=(cv::Mat const&) (mat.hpp:287)
> ==12142==    by 0x804E8A2: thread_main(void*) (threads.cpp:408)
> ==12142==    by 0x45A4D4B: start_thread (pthread_create.c:308)
> ==12142==    by 0x46A7DDD: clone (clone.S:130)

I really need help about it because I feel that probably exists an easy solution that I am not being able to see.

---

Some research:

-I searched until exhaustion inside the cv::Mat declaration in OpenCV documentation for a function that could do the trick but I had no luck. I can't find a function that without creating a new Mat might get the image (from a vector, char* or file). mat documentation

-Here is a similar question to what I need (with very unsatisfactory answers): link

Thanks in advance.

Answer 1

You cannot use calloc in the way you are. It will only allocate memory for the thread_data object but will not call its constructor or the constructors of its subobjects.

Change the following...

thread_data *tdata;
// declare array of structs
tdata = (thread_data * ) calloc(nthreads, sizeof(thread_data));

to...

thread_data *tdata = new thread_data();

I highly recommend using smart pointers such as std::unique_ptr so that you do not have to worry about calling delete .

std::unique_ptr<thread_data> tdata(new thread_data());