Hello,
I was modiying the NVTranscoder project from the Video_Codec_SDK_8.0.14 in order to adding some signal processing works into the video frames.
However, I encounter some problems when I turn the GPUMat into CUdeviceptr.
I was wondering how can I turn the GPUMat into CUdeviceptr. After I performed the blurring function, I want to turn the processed mat into a CUdeviceptr.
Besides, can anyone point out some problems of my implementation?
The code is as follows:
int main(int argc, char* argv[])
{
#if defined(WIN32) || defined(_WIN32) || defined(WIN64) || defined(_WIN64)
typedef HMODULE CUDADRIVER;
#else
typedef void *CUDADRIVER;
#endif
CUDADRIVER hHandleDriver = 0;
__cu(cuInit(0, __CUDA_API_VERSION, hHandleDriver));
__cu(cuvidInit(0));
EncodeConfig encodeConfig = { 0 };
encodeConfig.endFrameIdx = INT_MAX;
encodeConfig.bitrate = 5000000;
encodeConfig.rcMode = NV_ENC_PARAMS_RC_CONSTQP;
encodeConfig.gopLength = NVENC_INFINITE_GOPLENGTH;
encodeConfig.codec = NV_ENC_H264;
encodeConfig.fps = 0;
encodeConfig.qp = 28;
encodeConfig.i_quant_factor = DEFAULT_I_QFACTOR;
encodeConfig.b_quant_factor = DEFAULT_B_QFACTOR;
encodeConfig.i_quant_offset = DEFAULT_I_QOFFSET;
encodeConfig.b_quant_offset = DEFAULT_B_QOFFSET;
encodeConfig.presetGUID = NV_ENC_PRESET_DEFAULT_GUID;
encodeConfig.pictureStruct = NV_ENC_PIC_STRUCT_FRAME;
NVENCSTATUS nvStatus = CNvHWEncoder::ParseArguments(&encodeConfig, argc, argv);
if (nvStatus != NV_ENC_SUCCESS)
{
PrintHelp();
return 1;
}
if (!encodeConfig.inputFileName || !encodeConfig.outputFileName)
{
PrintHelp();
return 1;
}
encodeConfig.fOutput = fopen(encodeConfig.outputFileName, "wb");
if (encodeConfig.fOutput == NULL)
{
PRINTERR("Failed to create \"%s\"\n", encodeConfig.outputFileName);
return 1;
}
//init cuda
CUcontext cudaCtx;
CUdevice device;
__cu(cuDeviceGet(&device, encodeConfig.deviceID));
__cu(cuCtxCreate(&cudaCtx, CU_CTX_SCHED_AUTO, device));
CUcontext curCtx;
CUvideoctxlock ctxLock;
__cu(cuCtxPopCurrent(&curCtx));
__cu(cuvidCtxLockCreate(&ctxLock, curCtx));
CudaDecoder* pDecoder = new CudaDecoder;
FrameQueue* pFrameQueue = new CUVIDFrameQueue(ctxLock);
pDecoder->InitVideoDecoder(encodeConfig.inputFileName, ctxLock, pFrameQueue, encodeConfig.width, encodeConfig.height);
int decodedW, decodedH, decodedFRN, decodedFRD, isProgressive;
pDecoder->GetCodecParam(&decodedW, &decodedH, &decodedFRN, &decodedFRD, &isProgressive);
if (decodedFRN <= 0 || decodedFRD <= 0) {
decodedFRN = 30;
decodedFRD = 1;
}
if(encodeConfig.width <= 0 || encodeConfig.height <= 0) {
encodeConfig.width = decodedW;
encodeConfig.height = decodedH;
}
float fpsRatio = 1.f;
if (encodeConfig.fps <= 0) {
encodeConfig.fps = decodedFRN / decodedFRD;
}
else {
fpsRatio = (float)encodeConfig.fps * decodedFRD / decodedFRN;
}
encodeConfig.pictureStruct = (isProgressive ? NV_ENC_PIC_STRUCT_FRAME : 0);
pFrameQueue->init(encodeConfig.width, encodeConfig.height);
VideoEncoder* pEncoder = new VideoEncoder(ctxLock);
assert(pEncoder->GetHWEncoder());
nvStatus = pEncoder->GetHWEncoder()->Initialize(cudaCtx, NV_ENC_DEVICE_TYPE_CUDA);
if (nvStatus != NV_ENC_SUCCESS)
return 1;
encodeConfig.presetGUID = pEncoder->GetHWEncoder()->GetPresetGUID(encodeConfig.encoderPreset, encodeConfig.codec);
printf("Encoding input : \"%s\"\n", encodeConfig.inputFileName);
printf(" output : \"%s\"\n", encodeConfig.outputFileName);
printf(" codec : \"%s\"\n", encodeConfig.codec == NV_ENC_HEVC ? "HEVC" : "H264");
printf(" size : %dx%d\n", encodeConfig.width, encodeConfig.height);
printf(" bitrate : %d bits/sec\n", encodeConfig.bitrate);
printf(" vbvMaxBitrate : %d bits/sec\n", encodeConfig.vbvMaxBitrate);
printf(" vbvSize : %d bits\n", encodeConfig.vbvSize);
printf(" fps : %d frames/sec\n", encodeConfig.fps);
printf(" rcMode : %s\n", encodeConfig.rcMode == NV_ENC_PARAMS_RC_CONSTQP ? "CONSTQP" :
encodeConfig.rcMode == NV_ENC_PARAMS_RC_VBR ? "VBR" :
encodeConfig.rcMode == NV_ENC_PARAMS_RC_CBR ? "CBR" :
encodeConfig.rcMode == NV_ENC_PARAMS_RC_VBR_MINQP ? "VBR MINQP (deprecated)" :
encodeConfig.rcMode == NV_ENC_PARAMS_RC_CBR_LOWDELAY_HQ ? "CBR_LOWDELAY_HQ" :
encodeConfig.rcMode == NV_ENC_PARAMS_RC_CBR_HQ ? "CBR_HQ" :
encodeConfig.rcMode == NV_ENC_PARAMS_RC_VBR_HQ ? "VBR_HQ" : "UNKNOWN");
if (encodeConfig.gopLength == NVENC_INFINITE_GOPLENGTH)
printf(" goplength : INFINITE GOP \n");
else
printf(" goplength : %d \n", encodeConfig.gopLength);
printf(" B frames : %d \n", encodeConfig.numB);
printf(" QP : %d \n", encodeConfig.qp);
printf(" preset : %s\n", (encodeConfig.presetGUID == NV_ENC_PRESET_LOW_LATENCY_HQ_GUID) ? "LOW_LATENCY_HQ" :
(encodeConfig.presetGUID == NV_ENC_PRESET_LOW_LATENCY_HP_GUID) ? "LOW_LATENCY_HP" :
(encodeConfig.presetGUID == NV_ENC_PRESET_HQ_GUID) ? "HQ_PRESET" :
(encodeConfig.presetGUID == NV_ENC_PRESET_HP_GUID) ? "HP_PRESET" :
(encodeConfig.presetGUID == NV_ENC_PRESET_LOSSLESS_HP_GUID) ? "LOSSLESS_HP" : "LOW_LATENCY_DEFAULT");
printf("\n");
nvStatus = pEncoder->GetHWEncoder()->CreateEncoder(&encodeConfig);
if (nvStatus != NV_ENC_SUCCESS)
return 1;
nvStatus = pEncoder->AllocateIOBuffers(&encodeConfig);
if (nvStatus != NV_ENC_SUCCESS)
return 1;
unsigned long long lStart, lEnd, lFreq;
NvQueryPerformanceCounter(&lStart);
//start decoding thread
#ifdef _WIN32
HANDLE decodeThread = CreateThread(NULL, 0, DecodeProc, (LPVOID)pDecoder, 0, NULL);
#else
pthread_t pid;
pthread_create(&pid, NULL, DecodeProc, (void*)pDecoder);
#endif
//start encoding thread
int frmProcessed = 0;
int frmActual = 0;
while(!(pFrameQueue->isEndOfDecode() && pFrameQueue->isEmpty()) ) {
CUVIDPARSERDISPINFO pInfo;
if(pFrameQueue->dequeue(&pInfo)) {
CUdeviceptr dMappedFrame = 0;
unsigned int pitch;
CUVIDPROCPARAMS oVPP = { 0 };
oVPP.progressive_frame = pInfo.progressive_frame;
oVPP.second_field = 0;
oVPP.top_field_first = pInfo.top_field_first;
oVPP.unpaired_field = (pInfo.progressive_frame == 1 || pInfo.repeat_first_field <= 1);
cuvidMapVideoFrame(pDecoder->GetDecoder(), pInfo.picture_index, &dMappedFrame, &pitch, &oVPP);
[b]gpu::GpuMat gpuInput = gpu::GpuMat(decodedH, decodedW, CV_8UC3, (void*)dMappedFrame, pitch);
gpu::GpuMat d_dst(gpuInput.size(), gpuInput.type());
gpu::GpuMat d_buf;
gpu::GaussianBlur(gpuInput, d_dst, Size(3, 3), d_buf, 1);
cv::Mat result;
d_dst.download(result);[/b]
EncodeFrameConfig stEncodeConfig = { 0 };
NV_ENC_PIC_STRUCT picType = (pInfo.progressive_frame || pInfo.repeat_first_field >= 2 ? NV_ENC_PIC_STRUCT_FRAME :
(pInfo.top_field_first ? NV_ENC_PIC_STRUCT_FIELD_TOP_BOTTOM : NV_ENC_PIC_STRUCT_FIELD_BOTTOM_TOP));
<b>stEncodeConfig.dptr = result.data;</b>//dMappedFrame;
stEncodeConfig.pitch = pitch;
stEncodeConfig.width = encodeConfig.width;
stEncodeConfig.height = encodeConfig.height;
int dropOrDuplicate = MatchFPS(fpsRatio, frmProcessed, frmActual);
for (int i = 0; i <= dropOrDuplicate; i++) {
pEncoder->EncodeFrame(&stEncodeConfig, picType);
frmActual++;
}
frmProcessed++;
cuvidUnmapVideoFrame(pDecoder->GetDecoder(), dMappedFrame);
pFrameQueue->releaseFrame(&pInfo);
}
}
pEncoder->EncodeFrame(NULL, NV_ENC_PIC_STRUCT_FRAME, true);
#ifdef _WIN32
WaitForSingleObject(decodeThread, INFINITE);
#else
pthread_join(pid, NULL);
#endif
if (pEncoder->GetEncodedFrames() > 0)
{
NvQueryPerformanceCounter(&lEnd);
NvQueryPerformanceFrequency(&lFreq);
double elapsedTime = (double)(lEnd - lStart)/(double)lFreq;
printf("Total time: %fms, Decoded Frames: %d, Encoded Frames: %d, Average FPS: %f\n",
elapsedTime * 1000,
pDecoder->m_decodedFrames,
pEncoder->GetEncodedFrames(),
(float)pEncoder->GetEncodedFrames() / elapsedTime);
}
pEncoder->Deinitialize();
delete pDecoder;
delete pEncoder;
delete pFrameQueue;
cuvidCtxLockDestroy(ctxLock);
__cu(cuCtxDestroy(cudaCtx));
return 0;
}