MFVideoFormat_H264可以直接支持MFVideoFormat_RGB32的输入媒体类型吗？

Question

我正在写一个基于D3D11 + MediaFoundation的mp4(H264 + AAC)录音程序。基本流程是线程A使用D3D11API生成各种图形，然后将swap chain的back buffer复制到共享ID3D11Texture2D中，而记录线程B作为消费者将共享纹理复制到本地_capture_texture2d 并将其推入 MediaFoundation 的视频帧。
我有一个问题： https://docs.microsoft.com/en-us/windows/win32/medfound/h-264-video-encoder 要求 "输入媒体类型必须具有以下子类型之一：

##MFVideoFormat_I420

##MFVideoFormat_IYUV

##MFVideoFormat_NV12

##MFVideoFormat_YUY2

##MFVideoFormat_YV12"

但实际上为了和GDI/D2D互操作，我使用DXGI_FORMAT_B8G8R8A8_UNORM的格式 CreateSwapChain，对应MFVideoFormat_RGB32。在我的Win10和公司的其他几台win7/10机器上都很好用，不知道在用户真机环境下能不能正常使用？

所以我不确定是否必须将格式从MFVideoFormat_RGB32转换为MFVideoFormat_NV12？以及如何进行这种转换？用户的win7/8/10系统可以内置支持转换吗？（另外要求是尽量使用硬件加速）

HRESULT InitVideoStreamHelper(IMFSinkWriter *ppWriter, DWORD *pStreamIndex, const UINT32 uiWidth, const UINT32 uiHeight, const UINT32& fps, IMFDXGIDeviceManager* dxgi_device_manager)
{
     HRESULT hr = S_OK;

     // Set the output media type.
     CComPtr<IMFMediaType>   pMediaTypeOut = NULL;
     hr = MFCreateMediaTypeWrapper(&pMediaTypeOut);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeOut->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Video);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeOut->SetGUID(MF_MT_SUBTYPE, MFVideoFormat_H264);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeOut->SetUINT32(MF_MT_AVG_BITRATE, 1000000);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeOut->SetUINT32(MF_MT_INTERLACE_MODE, MFVideoInterlace_Progressive);
     RETURN_ON_FAIL(hr);

     hr = MFSetAttributeSize(pMediaTypeOut, MF_MT_FRAME_SIZE, uiWidth, uiHeight);
     RETURN_ON_FAIL(hr);

     hr = MFSetAttributeRatio(pMediaTypeOut, MF_MT_FRAME_RATE, fps, 1);
     RETURN_ON_FAIL(hr);

     hr = MFSetAttributeRatio(pMediaTypeOut, MF_MT_PIXEL_ASPECT_RATIO, 1, 1);
     RETURN_ON_FAIL(hr);

     hr = ppWriter->AddStream(pMediaTypeOut, pStreamIndex);
     RETURN_ON_FAIL(hr);

     // Set the input media type.
     CComPtr<IMFMediaType>   pMediaTypeIn = NULL;
     hr = MFCreateMediaTypeWrapper(&pMediaTypeIn);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeIn->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Video);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeIn->SetGUID(MF_MT_SUBTYPE, MFVideoFormat_RGB32);
     RETURN_ON_FAIL(hr);

     hr = pMediaTypeIn->SetUINT32(MF_MT_INTERLACE_MODE, MFVideoInterlace_Progressive);
     RETURN_ON_FAIL(hr);

     hr = MFSetAttributeSize(pMediaTypeIn, MF_MT_FRAME_SIZE, uiWidth, uiHeight);
     RETURN_ON_FAIL(hr);

     hr = MFSetAttributeRatio(pMediaTypeIn, MF_MT_FRAME_RATE, fps, 1);
     RETURN_ON_FAIL(hr);

     hr = MFSetAttributeRatio(pMediaTypeIn, MF_MT_PIXEL_ASPECT_RATIO, 1, 1);
     RETURN_ON_FAIL(hr);

     hr = ppWriter->SetInputMediaType(*pStreamIndex, pMediaTypeIn, NULL);
     RETURN_ON_FAIL(hr);

 // Tell the sink _writer to start accepting data.
 #if 1
     hr = MFTRegisterLocalByCLSID(
         __uuidof(CColorConvertDMO),
         MFT_CATEGORY_VIDEO_PROCESSOR,
         L"",
         MFT_ENUM_FLAG_SYNCMFT,
         0,
         NULL,
         0,
         NULL
     );
 #endif
     RETURN_ON_FAIL(hr);

     return hr;
 }

 HRESULT CMF_MP4FileWriter::MakeVideoFrame(IMFSample** sample, ID3D11Texture2D* texture)
 {   
     if (!_d3d_device || !_capture_texture2d)
     {
         return E_NOTIMPL;
     }

     HRESULT hr = E_NOTIMPL;

     CComPtr<IMFMediaBuffer> media_buffer;
     if (IsWindows8OrGreater())
     {
         hr = MFCreateDXGISurfaceBufferWrapper(__uuidof(ID3D11Texture2D), _capture_texture2d, 0, FALSE, &media_buffer);

         CComPtr<IMF2DBuffer> twod_buffer;
         hr = media_buffer->QueryInterface(&twod_buffer);
         RETURN_ON_FAIL(hr);

         DWORD length = 0;
         hr = twod_buffer->GetContiguousLength(&length);
         RETURN_ON_FAIL(hr);

         hr = media_buffer->SetCurrentLength(length);
         RETURN_ON_FAIL(hr);
     }
     else
     {   
         CComPtr<ID3D11DeviceContext> immediate_context;
         _d3d_device->GetImmediateContext(&immediate_context);

         D3D11_TEXTURE2D_DESC desc;
         _capture_texture2d->GetDesc(&desc);

         UINT32 img_size = 0;
         GUID dst_format = MFVideoFormat_RGB32;
         hr = MFCalculateImageSize(dst_format, desc.Width, desc.Height, &img_size);
         RETURN_ON_FAIL(hr);

         LONG dst_stride = 0;
         hr = MFGetStrideForBitmapInfoHeader(dst_format.Data1, desc.Width, &dst_stride);
         RETURN_ON_FAIL(hr);

         D3D11_MAPPED_SUBRESOURCE resource;
         UINT subresource = D3D11CalcSubresource(0, 0, 0);
         hr = immediate_context->Map(_capture_texture2d, subresource, D3D11_MAP_READ, 0, &resource);
         RETURN_ON_FAIL(hr);

         hr = MFCreateMemoryBuffer(img_size, &media_buffer);
         if (SUCCEEDED(hr))
         {
             BYTE *dst_data = NULL;
             hr = media_buffer->Lock(&dst_data, NULL, NULL);
             if (SUCCEEDED(hr))
             {
                 if (dst_stride >= 0)
                 {
                     hr = MFCopyImage(dst_data, dst_stride, (BYTE*)resource.pData, resource.RowPitch, desc.Width * 4, desc.Height);
                 }
                 else
                 {
                     BYTE *src_ptr = (BYTE*)resource.pData;
                     BYTE *dst_ptr = dst_data - dst_stride * (desc.Height - 1);
                     size_t copy_bytes = min(resource.RowPitch, (UINT)-dst_stride);
                     for (size_t i = 0; i < desc.Height; i++)
                     {
                         memcpy(dst_ptr, src_ptr, copy_bytes);

                         src_ptr += resource.RowPitch;
                         dst_ptr += dst_stride; 
                     }
                 }

                 media_buffer->SetCurrentLength(img_size);

                 media_buffer->Unlock();
             }
         }

         immediate_context->Unmap(_capture_texture2d, subresource);
     }   
     RETURN_ON_FAIL(hr);

     hr = MFCreateSampleWrapper(sample);
     RETURN_ON_FAIL(hr);

     hr = (*sample)->AddBuffer(media_buffer);
     RETURN_ON_FAIL(hr);

     return hr;
 }

 HRESULT CMF_MP4FileWriter::WriteVideoFrame(IMFSample* sample)
 {
     if (sample && _writer)
     {
         _video_frame_count++;

         HRESULT hr = sample->SetSampleTime(_video_frame_count * 10000000 / _video_fps);
         RETURN_ON_FAIL(hr);

         hr = sample->SetSampleDuration(10000000 / _video_fps);  
         RETURN_ON_FAIL(hr);

         hr = _writer->WriteSample(_video_sink_stream, sample);
         RETURN_ON_FAIL(hr);
     }

     return S_OK;
 }

Answer 1

所有 H.264 编码器都能够接受 MFVideoFormat_NV12 输入。其中一些采用其他格式的帧。如果您需要转换，Video Processor MFT可以为您转换格式。

请参阅此处的使用示例：

MFVideoFormat_H264可以直接支持MFVideoFormat_RGB32的输入媒体类型吗？

Can MFVideoFormat_H264 directly support the input media type of MFVideoFormat_RGB32?

direct3d

ms-media-foundation

##MFVideoFormat_I420

##MFVideoFormat_IYUV

##MFVideoFormat_NV12

##MFVideoFormat_YUY2

##MFVideoFormat_YV12"