I'm using Desktop Duplication from Windows API.
Here is the code to access next frame and get the rectangle of pixels that have change from previous frame.
//
// Get next frame and write it into Data
//
_Success_(*Timeout == false && return == DUPL_RETURN_SUCCESS)
DUPL_RETURN DUPLICATIONMANAGER::GetFrame(_Out_ FRAME_DATA* Data, _Out_ bool* Timeout)
{
IDXGIResource* DesktopResource = nullptr;
DXGI_OUTDUPL_FRAME_INFO FrameInfo;
// Get new frame
HRESULT hr = m_DeskDupl->AcquireNextFrame(10000, &FrameInfo, &DesktopResource);
if (hr == DXGI_ERROR_WAIT_TIMEOUT)
{
*Timeout = true;
return DUPL_RETURN_SUCCESS;
}
*Timeout = false;
if (FAILED(hr))
{
return ProcessFailure(m_Device, L"Failed to acquire next frame in DUPLICATIONMANAGER", L"Error", hr, FrameInfoExpectedErrors);
}
// If still holding old frame, destroy it
if (m_AcquiredDesktopImage)
{
m_AcquiredDesktopImage->Release();
m_AcquiredDesktopImage = nullptr;
}
// QI for IDXGIResource
hr = DesktopResource->QueryInterface(__uuidof(ID3D11Texture2D), reinterpret_cast<void **>(&m_AcquiredDesktopImage));
DesktopResource->Release();
DesktopResource = nullptr;
if (FAILED(hr))
{
return ProcessFailure(nullptr, L"Failed to QI for ID3D11Texture2D from acquired IDXGIResource in DUPLICATIONMANAGER", L"Error", hr);
}
// Get metadata
if (FrameInfo.TotalMetadataBufferSize)
{
// Old buffer too small
if (FrameInfo.TotalMetadataBufferSize > m_MetaDataSize)
{
if (m_MetaDataBuffer)
{
delete [] m_MetaDataBuffer;
m_MetaDataBuffer = nullptr;
}
m_MetaDataBuffer = new (std::nothrow) BYTE[FrameInfo.TotalMetadataBufferSize];
if (!m_MetaDataBuffer)
{
m_MetaDataSize = 0;
Data->MoveCount = 0;
Data->DirtyCount = 0;
return ProcessFailure(nullptr, L"Failed to allocate memory for metadata in DUPLICATIONMANAGER", L"Error", E_OUTOFMEMORY);
}
m_MetaDataSize = FrameInfo.TotalMetadataBufferSize;
}
UINT BufSize = FrameInfo.TotalMetadataBufferSize;
// Get move rectangles
hr = m_DeskDupl->GetFrameMoveRects(BufSize, reinterpret_cast<DXGI_OUTDUPL_MOVE_RECT*>(m_MetaDataBuffer), &BufSize);
if (FAILED(hr))
{
Data->MoveCount = 0;
Data->DirtyCount = 0;
return ProcessFailure(nullptr, L"Failed to get frame move rects in DUPLICATIONMANAGER", L"Error", hr, FrameInfoExpectedErrors);
}
Data->MoveCount = BufSize / sizeof(DXGI_OUTDUPL_MOVE_RECT);
BYTE* DirtyRects = m_MetaDataBuffer + BufSize;
BufSize = FrameInfo.TotalMetadataBufferSize - BufSize;
// Get dirty rectangles
hr = m_DeskDupl->GetFrameDirtyRects(BufSize, reinterpret_cast<RECT*>(DirtyRects), &BufSize);
if (FAILED(hr))
{
Data->MoveCount = 0;
Data->DirtyCount = 0;
return ProcessFailure(nullptr, L"Failed to get frame dirty rects in DUPLICATIONMANAGER", L"Error", hr, FrameInfoExpectedErrors);
}
Data->DirtyCount = BufSize / sizeof(RECT);
Data->MetaData = m_MetaDataBuffer;
}
Data->Frame = m_AcquiredDesktopImage;
Data->FrameInfo = FrameInfo;
//Here I would like to access pixel data from Data->Frame. A buffer of RGBA pixel
return DUPL_RETURN_SUCCESS;
}
Here is Frame_Data structure
typedef struct _FRAME_DATA
{
ID3D11Texture2D* Frame;
DXGI_OUTDUPL_FRAME_INFO FrameInfo;
_Field_size_bytes_((MoveCount * sizeof(DXGI_OUTDUPL_MOVE_RECT)) + (DirtyCount * sizeof(RECT))) BYTE* MetaData;
UINT DirtyCount;
UINT MoveCount;
} FRAME_DATA;
Is it possible to access pixel buffer data that have been modified from Data->Frame
Here is my code to access data :
BYTE* DISPLAYMANAGER::GetImageData(ID3D11Texture2D* texture2D, D3D11_TEXTURE2D_DESC Desc)
{
if (texture2D != NULL)
{
D3D11_TEXTURE2D_DESC description;
texture2D->GetDesc(&description);
description.BindFlags = 0;
description.CPUAccessFlags = D3D11_CPU_ACCESS_READ | D3D11_CPU_ACCESS_WRITE;
description.Usage = D3D11_USAGE_STAGING;
description.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
ID3D11Texture2D* texTemp = NULL;
HRESULT hr = m_Device->CreateTexture2D(&description, NULL, &texTemp);
if (FAILED(hr))
{
if (texTemp)
{
texTemp->Release();
texTemp = NULL;
}
return NULL;
}
m_DeviceContext->CopyResource(texTemp, texture2D);
D3D11_MAPPED_SUBRESOURCE mapped;
unsigned int subresource = D3D11CalcSubresource(0, 0, 0);
hr = m_DeviceContext->Map(texTemp, subresource, D3D11_MAP_READ_WRITE, 0, &mapped);
if (FAILED(hr))
{
texTemp->Release();
texTemp = NULL;
return NULL;
}
unsigned char *captureData = new unsigned char[Desc.Width * Desc.Height * 4];
RtlZeroMemory(captureData, Desc.Width * Desc.Height * 4);
const int pitch = mapped.RowPitch;
unsigned char *source = static_cast<unsigned char*>(mapped.pData);
unsigned char *dest = captureData;
for (int i = 0; i < Desc.Height; i++) {
memcpy(captureData, source, Desc.Width * 4);
source += pitch;
captureData += Desc.Width * 4;
}
for (int i = 0; i < Desc.Width * Desc.Height * 4; i++) {
//trace(L"Pixel[%d] = %x\n", i, dest[i]);
}
m_DeviceContext->Unmap(texTemp, 0);
return dest;
}
else
return NULL;
}
Thank you for your help!
The textures you obtain via duplication API are not necessarily accessible for CPU, for individual pixel access. To read the texture data, you might need to create a mappable staging texture and copy the obtained texture there. Then doing the mapping you would get a pointer to actual data. Note that this is, in general, not a performance friendly operation.
You will find related information in other answers as well:
How to work with pixels using Direct2D:
For those times when you absolutely have to do CPU pixel manipulation but still want a substantial degree of acceleration, you can manage your own mappable D3D11 textures. For example, you can use staging textures if you want to asynchronously manipulate your texture resources from the CPU.
Transferring textures across adapters in DirectX 11:
... copies it to a staging resource (created on the same device) using ID3D11DeviceContext::CopyResource. I then map that staging resource with Read...
Related
I'm working on a WASAPI UWP audio application with cpp/winrt which needs to take audio from an input and send it to an output after being processed.
I want to set my audio thread characteristics with AvSetMmThreadCharacteristicsW(L"Pro Audio", &taskIndex), but I just noticed this function (and most of avrt.h) is limited to WINAPI_PARTITION_DESKTOP and WINAPI_PARTITION_GAMES.
I think I need this because when my code is integrated into my UWP app, the audio input is full of discontinuity, and I have no issue in my test code which uses the avrt API.
Is there another way to configure my thread for audio processing?
Edit: here is my test program https://github.com/loics2/test-wasapi. The interesting part happens in the AudioStream class. I can't share my UWP app, but I can copy as is these classes into a Windows Runtime Component.
Edit 2: here's the audio thread code :
void AudioStream::StreamWorker()
{
WAVEFORMATEX* captureFormat = nullptr;
WAVEFORMATEX* renderFormat = nullptr;
RingBuffer<float> captureBuffer;
RingBuffer<float> renderBuffer;
BYTE* streamBuffer = nullptr;
unsigned int streamBufferSize = 0;
unsigned int bufferFrameCount = 0;
unsigned int numFramesPadding = 0;
unsigned int inputBufferSize = 0;
unsigned int outputBufferSize = 0;
DWORD captureFlags = 0;
winrt::hresult hr = S_OK;
// m_inputClient is a winrt::com_ptr<IAudioClient3>
if (m_inputClient) {
hr = m_inputClient->GetMixFormat(&captureFormat);
// m_audioCaptureClient is a winrt::com_ptr<IAudioCaptureClient>
if (!m_audioCaptureClient) {
hr = m_inputClient->Initialize(
AUDCLNT_SHAREMODE_SHARED,
AUDCLNT_STREAMFLAGS_EVENTCALLBACK,
0,
0,
captureFormat,
nullptr);
hr = m_inputClient->GetService(__uuidof(IAudioCaptureClient), m_audioCaptureClient.put_void());
hr = m_inputClient->SetEventHandle(m_inputReadyEvent.get());
hr = m_inputClient->Reset();
hr = m_inputClient->Start();
}
}
hr = m_inputClient->GetBufferSize(&inputBufferSize);
// multiplying the buffer size by the number of channels
inputBufferSize *= 2;
// m_outputClient is a winrt::com_ptr<IAudioClient3>
if (m_outputClient) {
hr = m_outputClient->GetMixFormat(&renderFormat);
// m_audioRenderClientis a winrt::com_ptr<IAudioRenderClient>
if (!m_audioRenderClient) {
hr = m_outputClient->Initialize(
AUDCLNT_SHAREMODE_SHARED,
AUDCLNT_STREAMFLAGS_EVENTCALLBACK,
0,
0,
captureFormat,
nullptr);
hr = m_outputClient->GetService(__uuidof(IAudioRenderClient), m_audioRenderClient.put_void());
hr = m_outputClient->SetEventHandle(m_outputReadyEvent.get());
hr = m_outputClient->Reset();
hr = m_outputClient->Start();
}
}
hr = m_outputClient->GetBufferSize(&outputBufferSize);
// multiplying the buffer size by the number of channels
outputBufferSize *= 2;
while (m_isRunning)
{
// ===== INPUT =====
// waiting for the capture event
WaitForSingleObject(m_inputReadyEvent.get(), INFINITE);
// getting the input buffer data
hr = m_audioCaptureClient->GetNextPacketSize(&bufferFrameCount);
while (SUCCEEDED(hr) && bufferFrameCount > 0) {
m_audioCaptureClient->GetBuffer(&streamBuffer, &bufferFrameCount, &captureFlags, nullptr, nullptr);
if (bufferFrameCount != 0) {
captureBuffer.write(reinterpret_cast<float*>(streamBuffer), bufferFrameCount * 2);
hr = m_audioCaptureClient->ReleaseBuffer(bufferFrameCount);
if (FAILED(hr)) {
m_audioCaptureClient->ReleaseBuffer(0);
}
}
else
{
m_audioCaptureClient->ReleaseBuffer(0);
}
hr = m_audioCaptureClient->GetNextPacketSize(&bufferFrameCount);
}
// ===== CALLBACK =====
auto size = captureBuffer.size();
float* userInputData = (float*)calloc(size, sizeof(float));
float* userOutputData = (float*)calloc(size, sizeof(float));
captureBuffer.read(userInputData, size);
OnData(userInputData, userOutputData, size / 2, 2, 48000);
renderBuffer.write(userOutputData, size);
free(userInputData);
free(userOutputData);
// ===== OUTPUT =====
// waiting for the render event
WaitForSingleObject(m_outputReadyEvent.get(), INFINITE);
// getting information about the output buffer
hr = m_outputClient->GetBufferSize(&bufferFrameCount);
hr = m_outputClient->GetCurrentPadding(&numFramesPadding);
// adjust the frame count with the padding
bufferFrameCount -= numFramesPadding;
if (bufferFrameCount != 0) {
hr = m_audioRenderClient->GetBuffer(bufferFrameCount, &streamBuffer);
auto count = (bufferFrameCount * 2);
if (renderBuffer.read(reinterpret_cast<float*>(streamBuffer), count) < count) {
// captureBuffer is not full enough, we should fill the remainder with 0
}
hr = m_audioRenderClient->ReleaseBuffer(bufferFrameCount, 0);
if (FAILED(hr)) {
m_audioRenderClient->ReleaseBuffer(0, 0);
}
}
else
{
m_audioRenderClient->ReleaseBuffer(0, 0);
}
}
exit:
// Cleanup code
}
I removed the error handling code for clarity, most of it is :
if (FAILED(hr))
goto exit;
#IInspectable was right, there's something wrong with my code : the audio processing is done by a library which then calls callbacks with some results.
In my callback, I try to raise a winrt::event, but it sometimes takes more than 50ms. When it happens, it blocks the audio thread, and creates discontinuity...
I'm developing a software that should capture what is happening on the screen to make some processing. One of the requirements is that the software has to run at least 30FPS.
I've tried several options and I'm going to shopw you two, But none of them fully works to my needs:
(1) Using Direct X - The problem with this approach is that I cannot run it at 30 FPS (I get between 15 and 20):
DirectXScreenCapturer.h:
#include "IScreenCapturer.h"
#include <Wincodec.h> // we use WIC for saving images
#include <d3d9.h> // DirectX 9 header
#pragma comment(lib, "d3d9.lib") // link to DirectX 9 library
class DirectXScreenCapturer : public IScreenCapturer
{
public:
DirectXScreenCapturer();
~DirectXScreenCapturer();
virtual bool CaptureScreen(cv::Mat&) override;
private:
IDirect3D9* _d3d;
IDirect3DDevice9* _device;
IDirect3DSurface9* _surface;
cv::Mat _screen;
};
DirectXScreenCapturer.cpp:
#include "DirectXScreenCapturer.h"
DirectXScreenCapturer::DirectXScreenCapturer() : _d3d(NULL), _device(NULL), _surface(NULL)
{
HRESULT hr = S_OK;
D3DPRESENT_PARAMETERS parameters = { 0 };
D3DDISPLAYMODE mode;
D3DLOCKED_RECT rc;
LPBYTE *shots = nullptr;
// init D3D and get screen size
_d3d = Direct3DCreate9(D3D_SDK_VERSION);
_d3d->GetAdapterDisplayMode(D3DADAPTER_DEFAULT, &mode);
parameters.Windowed = TRUE;
parameters.BackBufferCount = 1;
parameters.BackBufferHeight = mode.Height;
parameters.BackBufferWidth = mode.Width;
parameters.SwapEffect = D3DSWAPEFFECT_DISCARD;
parameters.hDeviceWindow = NULL;
// create device & capture surface
hr = _d3d->CreateDevice(D3DADAPTER_DEFAULT, _D3DDEVTYPE::D3DDEVTYPE_HAL, NULL, D3DCREATE_HARDWARE_VERTEXPROCESSING | D3DCREATE_DISABLE_PSGP_THREADING | D3DCREATE_PUREDEVICE, ¶meters, &_device);
hr = _device->CreateOffscreenPlainSurface(mode.Width, mode.Height, D3DFMT_A8R8G8B8, D3DPOOL_SYSTEMMEM, &_surface, nullptr);
// compute the required buffer size
hr = _surface->LockRect(&rc, NULL, 0);
hr = _surface->UnlockRect();
// allocate screenshots buffers
_screen = cv::Mat(mode.Height, mode.Width, CV_8UC4, rc.pBits);
}
DirectXScreenCapturer::~DirectXScreenCapturer()
{
}
bool DirectXScreenCapturer::CaptureScreen(cv::Mat& result)
{
_device->GetFrontBufferData(0, _surface);
result = _screen;
return true;
}
(2) I've tried DXGI with Direct X. This solution works really well in Real Time, but fails to capture the screen when another application is running in full screen, so the software doesn't work if I'm watching movies, playign games, etc:
DXGIScreenCapturer.h
#include "IScreenCapturer.h"
#include <DXGI1_2.h>
#pragma comment(lib, "Dxgi.lib")
#include <D3D11.h>
#pragma comment(lib, "D3D11.lib")
class DXGIScreenCapturer : public IScreenCapturer
{
public:
DXGIScreenCapturer();
~DXGIScreenCapturer();
bool Init();
virtual bool CaptureScreen(cv::Mat&) override;
private:
ID3D11Device* _lDevice;
ID3D11DeviceContext* _lImmediateContext;
IDXGIOutputDuplication* _lDeskDupl;
ID3D11Texture2D* _lAcquiredDesktopImage;
DXGI_OUTPUT_DESC _lOutputDesc;
DXGI_OUTDUPL_DESC _lOutputDuplDesc;
D3D11_MAPPED_SUBRESOURCE _resource;
ID3D11Texture2D* currTexture;
cv::Mat _result;
};
DXGIScreenCapturer.cpp
#include "DXGIScreenCapturer.h"
DXGIScreenCapturer::DXGIScreenCapturer()
{
Init();
}
DXGIScreenCapturer::~DXGIScreenCapturer()
{
}
bool DXGIScreenCapturer::Init() {
// Feature levels supported
D3D_FEATURE_LEVEL gFeatureLevels[] = {
D3D_FEATURE_LEVEL_11_0,
D3D_FEATURE_LEVEL_10_1,
D3D_FEATURE_LEVEL_10_0,
D3D_FEATURE_LEVEL_9_1
};
UINT gNumFeatureLevels = ARRAYSIZE(gFeatureLevels);
D3D_FEATURE_LEVEL lFeatureLevel;
HRESULT hr(E_FAIL);
hr = D3D11CreateDevice(nullptr, D3D_DRIVER_TYPE_HARDWARE, nullptr, D3D11_CREATE_DEVICE_FLAG::D3D11_CREATE_DEVICE_SINGLETHREADED, gFeatureLevels, gNumFeatureLevels, D3D11_SDK_VERSION, &_lDevice, &lFeatureLevel, &_lImmediateContext);
if (FAILED(hr))
return false;
if (!_lDevice)
return false;
// Get DXGI device
IDXGIDevice* lDxgiDevice;
hr = _lDevice->QueryInterface(__uuidof(IDXGIDevice), reinterpret_cast<void**>(&lDxgiDevice));
if (FAILED(hr))
return false;
// Get DXGI adapter
IDXGIAdapter* lDxgiAdapter;
hr = lDxgiDevice->GetParent(__uuidof(IDXGIAdapter), reinterpret_cast<void**>(&lDxgiAdapter));
lDxgiDevice->Release();
lDxgiDevice = nullptr;
if (FAILED(hr))
return false;
UINT Output = 0;
// Get output
IDXGIOutput* lDxgiOutput;
hr = lDxgiAdapter->EnumOutputs(Output, &lDxgiOutput);
if (FAILED(hr))
return false;
lDxgiAdapter->Release();
lDxgiAdapter = nullptr;
hr = lDxgiOutput->GetDesc(&_lOutputDesc);
if (FAILED(hr))
return false;
// QI for Output 1
IDXGIOutput1* lDxgiOutput1;
hr = lDxgiOutput->QueryInterface(__uuidof(lDxgiOutput1), reinterpret_cast<void**>(&lDxgiOutput1));
lDxgiOutput->Release();
lDxgiOutput = nullptr;
if (FAILED(hr))
return false;
// Create desktop duplication
hr = lDxgiOutput1->DuplicateOutput(_lDevice, &_lDeskDupl);
if (FAILED(hr))
return false;
lDxgiOutput1->Release();
lDxgiOutput1 = nullptr;
// Create GUI drawing texture
_lDeskDupl->GetDesc(&_lOutputDuplDesc);
// Create CPU access texture
D3D11_TEXTURE2D_DESC desc;
desc.Width = _lOutputDuplDesc.ModeDesc.Width;
desc.Height = _lOutputDuplDesc.ModeDesc.Height;
desc.Format = _lOutputDuplDesc.ModeDesc.Format;
desc.ArraySize = 1;
desc.BindFlags = 0;
desc.MiscFlags = 0;
desc.SampleDesc.Count = 1;
desc.SampleDesc.Quality = 0;
desc.MipLevels = 1;
desc.CPUAccessFlags = D3D11_CPU_ACCESS_FLAG::D3D11_CPU_ACCESS_READ;
desc.Usage = D3D11_USAGE::D3D11_USAGE_STAGING;
hr = _lDevice->CreateTexture2D(&desc, NULL, &currTexture);
if (!currTexture)
{
hr = _lDeskDupl->ReleaseFrame();
return false;
}
while (!CaptureScreen(_result));
_result = cv::Mat(desc.Height, desc.Width, CV_8UC4, _resource.pData);
return true;
}
bool DXGIScreenCapturer::CaptureScreen(cv::Mat& output)
{
HRESULT hr(E_FAIL);
IDXGIResource* lDesktopResource = nullptr;
DXGI_OUTDUPL_FRAME_INFO lFrameInfo;
hr = _lDeskDupl->AcquireNextFrame(999, &lFrameInfo, &lDesktopResource);
if (FAILED(hr))
return false;
if (lFrameInfo.LastPresentTime.HighPart == 0) // not interested in just mouse updates, which can happen much faster than 60fps if you really shake the mouse
{
hr = _lDeskDupl->ReleaseFrame();
return false;
}
// QI for ID3D11Texture2D
hr = lDesktopResource->QueryInterface(__uuidof(ID3D11Texture2D), reinterpret_cast<void **>(&_lAcquiredDesktopImage));
lDesktopResource->Release();
lDesktopResource = nullptr;
if (FAILED(hr))
{
hr = _lDeskDupl->ReleaseFrame();
return false;
}
_lImmediateContext->CopyResource(currTexture, _lAcquiredDesktopImage);
UINT subresource = D3D11CalcSubresource(0, 0, 0);
_lImmediateContext->Map(currTexture, subresource, D3D11_MAP_READ, 0, &_resource);
_lImmediateContext->Unmap(currTexture, 0);
hr = _lDeskDupl->ReleaseFrame();
output = _result;
return true;
}
Please note that IScreenCapturer is just and interface to quickly swap implementations and also note that the return result is a cv::Mat object (OpenCV to do the rest of processing).
Any help on this issue? I'm still trying to figure out a solution that can capture the screen at least 30 FPS and that can capture the whole screen even when an app is running at full screen.
I'm currently trying to develop a personal project where I will light up LEDs according to what's happening on the screen of my computer.
I've tried several solutions to capture my screen, and DirectX11 with DXGI is the fastest way I found to have a good FPS Rate.
My only issue is the following: When in full screen (for example, watching Netflix trogh Win10 App or playing any game in fullscreen), it seems that nothing is captured. I have two functions (one to set up and another one to capture a frame:
Setup Up Function:
bool DXGIScreenCapturer::Init() {
int lresult(-1);
D3D_FEATURE_LEVEL lFeatureLevel;
HRESULT hr(E_FAIL);
hr = D3D11CreateDevice(
nullptr,
D3D_DRIVER_TYPE_HARDWARE,
nullptr,
0,
gFeatureLevels,
gNumFeatureLevels,
D3D11_SDK_VERSION,
&_lDevice,
&lFeatureLevel,
&_lImmediateContext);
if (FAILED(hr))
return false;
if (!_lDevice)
return false;
// Get DXGI device
ComPtr<IDXGIDevice> lDxgiDevice;
hr = _lDevice.As(&lDxgiDevice);
if (FAILED(hr))
return false;
// Get DXGI adapter
ComPtr<IDXGIAdapter> lDxgiAdapter;
hr = lDxgiDevice->GetParent(__uuidof(IDXGIAdapter), &lDxgiAdapter);
if (FAILED(hr))
return false;
lDxgiDevice.Reset();
UINT Output = 0;
// Get output
ComPtr<IDXGIOutput> lDxgiOutput;
hr = lDxgiAdapter->EnumOutputs(Output, &lDxgiOutput);
if (FAILED(hr))
return false;
lDxgiAdapter.Reset();
hr = lDxgiOutput->GetDesc(&_lOutputDesc);
if (FAILED(hr))
return false;
// QI for Output 1
ComPtr<IDXGIOutput1> lDxgiOutput1;
hr = lDxgiOutput.As(&lDxgiOutput1);
if (FAILED(hr))
return false;
lDxgiOutput.Reset();
// Create desktop duplication
hr = lDxgiOutput1->DuplicateOutput(_lDevice.Get(), &_lDeskDupl);
if (FAILED(hr))
return false;
lDxgiOutput1.Reset();
// Create GUI drawing texture
_lDeskDupl->GetDesc(&_lOutputDuplDesc);
// Create CPU access texture
_desc.Width = _lOutputDuplDesc.ModeDesc.Width;
_desc.Height = _lOutputDuplDesc.ModeDesc.Height;
_desc.Format = _lOutputDuplDesc.ModeDesc.Format;
std::cout << _desc.Width << "x" << _desc.Height << "\n\n\n";
_desc.ArraySize = 1;
_desc.BindFlags = 0;
_desc.MiscFlags = 0;
_desc.SampleDesc.Count = 1;
_desc.SampleDesc.Quality = 0;
_desc.MipLevels = 1;
_desc.CPUAccessFlags = D3D11_CPU_ACCESS_FLAG::D3D11_CPU_ACCESS_READ;
_desc.Usage = D3D11_USAGE::D3D11_USAGE_STAGING;
while (!CaptureScreen(_result));
_result = cv::Mat(_desc.Height, _desc.Width, CV_8UC4, _resource.pData);
return true;
}
And the capture function:
bool DXGIScreenCapturer::CaptureScreen(cv::Mat& output)
{
HRESULT hr(E_FAIL);
ComPtr<IDXGIResource> lDesktopResource = nullptr;
DXGI_OUTDUPL_FRAME_INFO lFrameInfo;
ID3D11Texture2D* currTexture = NULL;
hr = _lDeskDupl->AcquireNextFrame(999, &lFrameInfo, &lDesktopResource);
if (FAILED(hr))
return false;
if (lFrameInfo.LastPresentTime.HighPart == 0) // not interested in just mouse updates, which can happen much faster than 60fps if you really shake the mouse
{
hr = _lDeskDupl->ReleaseFrame();
return false;
}
//int accum_frames = lFrameInfo.AccumulatedFrames;
//if (accum_frames > 1) {// && current_frame != 1) {
// // TOO MANY OF THESE is the problem
// // especially after having to wait >17ms in AcquireNextFrame()
//}
// QI for ID3D11Texture2D
hr = lDesktopResource.As(&_lAcquiredDesktopImage);
// Copy image into a newly created CPU access texture
hr = _lDevice->CreateTexture2D(&_desc, NULL, &currTexture);
if (FAILED(hr))
{
hr = _lDeskDupl->ReleaseFrame();
return false;
}
if (!currTexture)
{
hr = _lDeskDupl->ReleaseFrame();
return false;
}
_lImmediateContext->CopyResource(currTexture, _lAcquiredDesktopImage.Get());
UINT subresource = D3D11CalcSubresource(0, 0, 0);
_lImmediateContext->Map(currTexture, subresource, D3D11_MAP_READ, 0, &_resource);
_lImmediateContext->Unmap(currTexture, 0);
currTexture->Release();
hr = _lDeskDupl->ReleaseFrame();
output = _result;
return true;
}
Copying a ID3D11Texture2D to a byte array is causing out of memory error c++
The copy is happening every frame and it runs for around 5 seconds until I get the memory error, which is happening on either
HRESULT hr = device->CreateTexture2D(&description, NULL, &texTemp);
Result in GetImageData returning NULL
OR
when a new byte array is created at
BYTE* dest = new BYTE[(*nWidth)*(*nHeight) * 4];
Giving,
Unhandled exception at 0x76414DBD in NDIlib_Send_Video.x86.exe: Microsoft C++ exception: std::bad_alloc at memory location 0x0107F5D8
I'm not sure which memory I should be releasing, probably the byte arrays.., I've tried this using delete[] aByteArray, but this causes a crash.
Here's my function for doing the copy,
BYTE* GetImageData(ID3D11Device* device, ID3D11DeviceContext* context, ID3D11Texture2D* texture, /*OUT*/ int* nWidth, /*OUT*/ int* nHeight) {
if (texture) {
D3D11_TEXTURE2D_DESC description;
texture->GetDesc(&description);
description.BindFlags = 0;
description.CPUAccessFlags = D3D11_CPU_ACCESS_READ | D3D11_CPU_ACCESS_WRITE;
description.Usage = D3D11_USAGE_STAGING;
ID3D11Texture2D* texTemp = NULL;
HRESULT hr = device->CreateTexture2D(&description, NULL, &texTemp);
if (FAILED(hr))
{
if (hr == E_OUTOFMEMORY) {
printf("GetImageData - CreateTexture2D - OUT OF MEMORY \n");
}
if (texTemp)
{
texTemp->Release();
texTemp = NULL;
}
return NULL;
}
context->CopyResource(texTemp, texture);
D3D11_MAPPED_SUBRESOURCE mapped;
unsigned int subresource = 0;
hr = context->Map(texTemp, 0, D3D11_MAP_READ, 0, &mapped);
if (FAILED(hr))
{
printf("GetImageData - Map - FAILED \n");
texTemp->Release();
texTemp = NULL;
return NULL;
}
*nWidth = description.Width;
*nHeight = description.Height;
const int pitch = mapped.RowPitch;
BYTE* source = (BYTE*)(mapped.pData);
BYTE* dest = new BYTE[(*nWidth)*(*nHeight) * 4];
BYTE* destTemp = dest;
for (int i = 0; i < *nHeight; ++i)
{
memcpy(destTemp, source, *nWidth * 4);
source += pitch;
destTemp += *nWidth * 4;
}
context->Unmap(texTemp, 0);
return dest;
}
else {
printf("GetImageData - texture null - FAILED \n");
return NULL;
}
}
And here's how I'm using GetImageData from my main function
BYTE* p_frame = (BYTE*)GetImageData(g_d3dDevice, g_d3dDeviceContext, g_pDeskTexture, &w, &h);
if (p_frame == NULL) printf("GetImageData - Returned NULL \n");
else {
printf("GetImageData - OK \n");
}
Where do you delete dest* or in the case after being returned p_frame*? You are allocating data using new and returning this pointer, if you don't delete this you will eventually run out of memory, especially if you're calling this every frame.
You should also consider some form of memory management and define ownership of what's being allocated, passing back raw pointers is a prime suspect for memory leaks.
I'm trying to split up an contiguous buffer into 3 byte channels (RGB). Here is my acutal workflow to get the buffer filled with an image:
Set up an Source Reader (MFVideoFormat_RGB32)
Receive video format information
Read first image and convert to contiguous buffer...
In addition to that, here is the code:
HRESULT hr = S_OK;
IMFAttributes *attributes = NULL;
SafeRelease(&_sourcereader);
hr = MFCreateAttributes(&attributes, 1);
if (FAILED(hr)) {
// TODO: set error
return false;
}
hr = attributes->SetUINT32(MF_SOURCE_READER_ENABLE_VIDEO_PROCESSING, true);
if (FAILED(hr)) {
// TODO: set error
return false;
}
// conversion from qstring to const wchar*
const WCHAR* wfilename = filename.toStdWString().c_str();
// create source reader from file with attributes
hr = MFCreateSourceReaderFromURL(wfilename, attributes, &_sourcereader);
if (FAILED(hr)) {
// TODO: set error
return false;
}
// configure sourcereader for progressive RGB32 frames
IMFMediaType *mediatype = NULL;
hr = MFCreateMediaType(&mediatype);
if (SUCCEEDED(hr))
{
hr = mediatype->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Video);
}
if (SUCCEEDED(hr))
{
hr = mediatype->SetGUID(MF_MT_SUBTYPE, MFVideoFormat_RGB32);
}
if (SUCCEEDED(hr))
{
hr = _sourcereader->SetCurrentMediaType(
(DWORD)MF_SOURCE_READER_FIRST_VIDEO_STREAM,
NULL, mediatype);
}
// Ensure the stream is selected.
if (SUCCEEDED(hr))
{
hr = _sourcereader->SetStreamSelection(
(DWORD)MF_SOURCE_READER_FIRST_VIDEO_STREAM, TRUE);
}
if (FAILED(hr)) {
// TODO: Error log for failed configuration
std::cout << "(ConfigureSourceReader) Configuration failed" << std::endl;
return false;
}
//------------------------------------------------------------------
//---------------------- Get Video Format Infos --------------------
//------------------------------------------------------------------
GUID subtype = { 0 };
// Get the media type from the stream.
hr = _sourcereader->GetCurrentMediaType(
(DWORD)MF_SOURCE_READER_FIRST_VIDEO_STREAM, &mediatype );
// Make sure it is a video format.
hr = mediatype->GetGUID(MF_MT_SUBTYPE, &subtype);
if (subtype != MFVideoFormat_RGB32)
{
hr = E_UNEXPECTED;
// TODO: Error log message
SafeRelease(&mediatype);
return false;
}
//------------------------------------------------------------------
// Get the width and height
UINT32 width = 0, height = 0;
hr = MFGetAttributeSize(mediatype, MF_MT_FRAME_SIZE, &width, &height);
if (FAILED(hr))
{
// TODO: Error log message
SafeRelease(&mediatype);
return false;
}
//assign dimensions to VideoInfo
_videoinfo.imageHeight = height; _videoinfo.imageWidth = width;
//std::cout << "(GetVideoFormat) width: " << width << ", height: " << height << std::endl;
//------------------------------------------------------------------
//get framerate
UINT32 framerate_num = 0, framerate_denom = 0;
hr = MFGetAttributeRatio(mediatype, MF_MT_FRAME_RATE, &framerate_num, &framerate_denom);
if (FAILED(hr))
{
// TODO: Error log message
SafeRelease(&mediatype);
return false;
}
//set frame rate in struct
_videoinfo.fps = framerate_num / framerate_denom; // TODO: check for valid fps 24,25,30 ...
//------------------------------------------------------------------
// Get length
LONGLONG length = 0;
PROPVARIANT var;
PropVariantInit(&var);
hr = _sourcereader->GetPresentationAttribute((DWORD)MF_SOURCE_READER_MEDIASOURCE,
MF_PD_DURATION,
&var
);
if (SUCCEEDED(hr)) {
assert(var.vt == VT_UI8);
length = var.hVal.QuadPart;
} else {
// TODO : erro log msg
return false;
}
//Get total framenumber and length: save to info struct
_videoinfo.noofFrames = length / 10000000 * this->getFrameRate(); // incl. conversion from nano sec to sec
_videoinfo.duration = length;
//------------------------------------------------------------------
// Get the stride to find out if the bitmap is top-down or bottom-up.
LONG lStride = 0;
lStride = (LONG)MFGetAttributeUINT32(mediatype, MF_MT_DEFAULT_STRIDE, 1);
_videoinfo.stride = lStride;
_videoinfo.bTopDown = (lStride > 0);
//------------------------------------------------------------------
SafeRelease(&mediatype);
// return true and flag if initialization went well
_bInitialized = true;
return true;
After that I call a function to read a single frame (at the moment the first one).
HRESULT hr = S_OK;
IMFSample *pSample = NULL;
IMFMediaBuffer *buffer = NULL;
DWORD streamIndex, flags;
LONGLONG llTimeStamp;
// Read Sample (RGB32)
hr = _sourcereader->ReadSample (
(DWORD) MF_SOURCE_READER_FIRST_VIDEO_STREAM,
0,
&streamIndex,
&flags,
&llTimeStamp,
&pSample);
if (FAILED (hr)) {
// TODO handle fail case
}
//convert sample data to buffer
hr = pSample->ConvertToContiguousBuffer(&buffer);
if (FAILED (hr)) {
// TODO handle fail case
}
I know that by calling the function buffer->Lock(&pixels, NULL, &nPixels) that I can get the BYTE-stream stored in pixels. In my case I create a custom image with the given height and width (from SourceReader; [first function]). From the empty image I can get an empty color matrix which has to be filled with the following funtion: Color (byte red, byte green, byte blue)
I dont know how to split my RGB32 BYTE array into the single channels to fill my image? Maybe it is a silly question but I am relatively new to this area...
For RGB32 the byte format is:
R=Red
G=Green
B=Blue
A=Transparency
RGBARGBARGBA...
A very simple pseudo-code example of extracting the channels is shown below.
for (int row = 0; row < height; row++) {
for (int col = 0; col < stride; col += 4) {
redBuf[rIndex++] = sample[row * stride + col];
greenBuf[gIndex++] = sample[row * stride + col + 1];
blueBuf[bIndex++] = sample[row * stride + col + 2];
transparencyBuf[tIndex++] = sample[row * stride + col + 3];
}
}