C++ InputArray::getGpuMat方法代码示例

void cv::cuda::mulAndScaleSpectrums(InputArray _src1, InputArray _src2, OutputArray _dst, int flags, float scale, bool conjB, Stream& stream)
{
#ifndef HAVE_CUFFT
    (void) _src1;
    (void) _src2;
    (void) _dst;
    (void) flags;
    (void) scale;
    (void) conjB;
    (void) stream;
    throw_no_cuda();
#else
    (void)flags;

    typedef void (*Caller)(const PtrStep<cufftComplex>, const PtrStep<cufftComplex>, float scale, PtrStepSz<cufftComplex>, cudaStream_t stream);
    static Caller callers[] = { device::mulAndScaleSpectrums, device::mulAndScaleSpectrums_CONJ };

    GpuMat src1 = _src1.getGpuMat();
    GpuMat src2 = _src2.getGpuMat();

    CV_Assert( src1.type() == src2.type() && src1.type() == CV_32FC2);
    CV_Assert( src1.size() == src2.size() );

    _dst.create(src1.size(), CV_32FC2);
    GpuMat dst = _dst.getGpuMat();

    Caller caller = callers[(int)conjB];
    caller(src1, src2, scale, dst, StreamAccessor::getStream(stream));
#endif
}

void cv::cuda::divide(InputArray _src1, InputArray _src2, OutputArray _dst, double scale, int dtype, Stream& stream)
{
    if (_src1.type() == CV_8UC4 && _src2.type() == CV_32FC1)
    {
        GpuMat src1 = _src1.getGpuMat();
        GpuMat src2 = _src2.getGpuMat();

        CV_Assert( src1.size() == src2.size() );

        _dst.create(src1.size(), src1.type());
        GpuMat dst = _dst.getGpuMat();

        divMat_8uc4_32f(src1, src2, dst, stream);
    }
    else if (_src1.type() == CV_16SC4 && _src2.type() == CV_32FC1)
    {
        GpuMat src1 = _src1.getGpuMat();
        GpuMat src2 = _src2.getGpuMat();

        CV_Assert( src1.size() == src2.size() );

        _dst.create(src1.size(), src1.type());
        GpuMat dst = _dst.getGpuMat();

        divMat_16sc4_32f(src1, src2, dst, stream);
    }
    else
    {
        arithm_op(_src1, _src2, _dst, GpuMat(), scale, dtype, stream, divMat, divScalar);
    }
}

/**
 * Allocate memeroy for all the buffers on GPU
 */
void DepthmapDenoiseWeightedHuberImpl::allocate(int _rows, int _cols, InputArray _gxin, InputArray _gyin){
    const GpuMat& gxin = _gxin.getGpuMat();
    const GpuMat& gyin = _gyin.getGpuMat();
    
    rows = _rows;
    cols = _cols;
    if(!(rows % 32 == 0 && cols % 32 == 0 && cols >= 64)){
        CV_Assert(!"For performance reasons, DepthmapDenoiseWeightedHuber currenty only supports multiple of 32 image sizes with cols >= 64. Pad the image to achieve this.");
    }
    

    if(!_a.data){
        _a.create(1, rows*cols, CV_32FC1);
        _a = _a.reshape(0, rows);
    }

    FLATALLOC(_d, _a);

    cachedG = 1;
    if(gxin.empty() || gyin.empty()){
        if(gxin.empty()){
            FLATALLOC(_gx, _d);
            cachedG = 0;
        }else{
            _gx = gxin;
        }

        if(gyin.empty()){
            FLATALLOC(_gy, _d);
            cachedG = 0;
        }else{
            _gy = gyin;
        }
    }else{
        
        if(!gxin.isContinuous()){
            FLATALLOC(_gx, _d);
//             gxin.copyTo(_gx,cvStream);
            cvStream.enqueueCopy(gxin, _gx);
        }
        if(!gyin.isContinuous()){
            FLATALLOC(_gy, _d);
//             gyin.copyTo(_gy,cvStream);
            cvStream.enqueueCopy(gyin, _gy);
        }
    }

    FLATALLOC(_qx, _d);
    FLATALLOC(_qy, _d);
    FLATALLOC(_g1, _d);
    FLATALLOC(stableDepth, _d);
    memZero(_qx, cvStream);
    memZero(_qy, cvStream);
    alloced = 1;
}

void cv::softcascade::SCascade::detect(InputArray _image, InputArray _rois, OutputArray _objects, cv::gpu::Stream& s) const
{
    CV_Assert(fields);

    // only color images and precomputed integrals are supported
    int type = _image.type();
    CV_Assert(type == CV_8UC3 || type == CV_32SC1 || (!_rois.empty()));

    const cv::gpu::GpuMat image = _image.getGpuMat();

    if (_objects.empty()) _objects.create(1, 4096 * sizeof(Detection), CV_8UC1);

    cv::gpu::GpuMat rois = _rois.getGpuMat(), objects = _objects.getGpuMat();

    /// roi
    Fields& flds = *fields;
    int shr = flds.shrinkage;

    flds.mask.create( rois.cols / shr, rois.rows / shr, rois.type());

    device::shrink(rois, flds.mask);
    //cv::gpu::transpose(flds.genRoiTmp, flds.mask, s);

    if (type == CV_8UC3)
    {
        flds.update(image.rows, image.cols, flds.shrinkage);

        if (flds.check((float)minScale, (float)maxScale, scales))
            flds.createLevels(image.rows, image.cols);

        flds.preprocessor->apply(image, flds.shrunk);
        integral(flds.shrunk, flds.hogluv, flds.integralBuffer, s);
    }
    else
    {
        if (s)
            s.enqueueCopy(image, flds.hogluv);
        else
            image.copyTo(flds.hogluv);
    }

    flds.detect(objects, s);

    if ( (flags && NMS_MASK) != NO_REJECT)
    {
        cv::gpu::GpuMat spr(objects, cv::Rect(0, 0, flds.suppressed.cols, flds.suppressed.rows));
        flds.suppress(objects, s);
        flds.suppressed.copyTo(spr);
    }
}

void cv::cuda::bilateralFilter(InputArray _src, OutputArray _dst, int kernel_size, float sigma_color, float sigma_spatial, int borderMode, Stream& stream)
{
    using cv::cuda::device::imgproc::bilateral_filter_gpu;

    typedef void (*func_t)(const PtrStepSzb& src, PtrStepSzb dst, int kernel_size, float sigma_spatial, float sigma_color, int borderMode, cudaStream_t s);

    static const func_t funcs[6][4] =
    {
        {bilateral_filter_gpu<uchar>      , 0 /*bilateral_filter_gpu<uchar2>*/ , bilateral_filter_gpu<uchar3>      , bilateral_filter_gpu<uchar4>      },
        {0 /*bilateral_filter_gpu<schar>*/, 0 /*bilateral_filter_gpu<schar2>*/ , 0 /*bilateral_filter_gpu<schar3>*/, 0 /*bilateral_filter_gpu<schar4>*/},
        {bilateral_filter_gpu<ushort>     , 0 /*bilateral_filter_gpu<ushort2>*/, bilateral_filter_gpu<ushort3>     , bilateral_filter_gpu<ushort4>     },
        {bilateral_filter_gpu<short>      , 0 /*bilateral_filter_gpu<short2>*/ , bilateral_filter_gpu<short3>      , bilateral_filter_gpu<short4>      },
        {0 /*bilateral_filter_gpu<int>*/  , 0 /*bilateral_filter_gpu<int2>*/   , 0 /*bilateral_filter_gpu<int3>*/  , 0 /*bilateral_filter_gpu<int4>*/  },
        {bilateral_filter_gpu<float>      , 0 /*bilateral_filter_gpu<float2>*/ , bilateral_filter_gpu<float3>      , bilateral_filter_gpu<float4>      }
    };

    sigma_color = (sigma_color <= 0 ) ? 1 : sigma_color;
    sigma_spatial = (sigma_spatial <= 0 ) ? 1 : sigma_spatial;

    int radius = (kernel_size <= 0) ? cvRound(sigma_spatial*1.5) : kernel_size/2;
    kernel_size = std::max(radius, 1)*2 + 1;

    GpuMat src = _src.getGpuMat();

    CV_Assert( src.depth() <= CV_32F && src.channels() <= 4 );
    CV_Assert( borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP );

    const func_t func = funcs[src.depth()][src.channels() - 1];
    CV_Assert( func != 0 );

    _dst.create(src.size(), src.type());
    GpuMat dst = _dst.getGpuMat();

    func(src, dst, kernel_size, sigma_spatial, sigma_color, borderMode, StreamAccessor::getStream(stream));
}

void cv::cuda::rotate(InputArray _src, OutputArray _dst, Size dsize, double angle, double xShift, double yShift, int interpolation, Stream& stream)
{
    typedef void (*func_t)(const GpuMat& src, GpuMat& dst, Size dsize, double angle, double xShift, double yShift, int interpolation, cudaStream_t stream);
    static const func_t funcs[6][4] =
    {
        {NppRotate<CV_8U, nppiRotate_8u_C1R>::call, 0, NppRotate<CV_8U, nppiRotate_8u_C3R>::call, NppRotate<CV_8U, nppiRotate_8u_C4R>::call},
        {0,0,0,0},
        {NppRotate<CV_16U, nppiRotate_16u_C1R>::call, 0, NppRotate<CV_16U, nppiRotate_16u_C3R>::call, NppRotate<CV_16U, nppiRotate_16u_C4R>::call},
        {0,0,0,0},
        {0,0,0,0},
        {NppRotate<CV_32F, nppiRotate_32f_C1R>::call, 0, NppRotate<CV_32F, nppiRotate_32f_C3R>::call, NppRotate<CV_32F, nppiRotate_32f_C4R>::call}
    };

    GpuMat src = _src.getGpuMat();

    CV_Assert( src.depth() == CV_8U || src.depth() == CV_16U || src.depth() == CV_32F );
    CV_Assert( src.channels() == 1 || src.channels() == 3 || src.channels() == 4 );
    CV_Assert( interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC );

    _dst.create(dsize, src.type());
    GpuMat dst = _dst.getGpuMat();

    dst.setTo(Scalar::all(0), stream);

    funcs[src.depth()][src.channels() - 1](src, dst, dsize, angle, xShift, yShift, interpolation, StreamAccessor::getStream(stream));
}

void cv::gpu::split(InputArray _src, std::vector<GpuMat>& dst, Stream& stream)
{
    GpuMat src = _src.getGpuMat();
    dst.resize(src.channels());
    if(src.channels() > 0)
        split_caller(src, &dst[0], stream);
}

void cv::gpu::pyrDown(InputArray _src, OutputArray _dst, Stream& stream)
{
    using namespace cv::gpu::cudev::imgproc;

    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
    static const func_t funcs[6][4] =
    {
        {pyrDown_gpu<uchar>      , 0 /*pyrDown_gpu<uchar2>*/ , pyrDown_gpu<uchar3>      , pyrDown_gpu<uchar4>      },
        {0 /*pyrDown_gpu<schar>*/, 0 /*pyrDown_gpu<schar2>*/ , 0 /*pyrDown_gpu<schar3>*/, 0 /*pyrDown_gpu<schar4>*/},
        {pyrDown_gpu<ushort>     , 0 /*pyrDown_gpu<ushort2>*/, pyrDown_gpu<ushort3>     , pyrDown_gpu<ushort4>     },
        {pyrDown_gpu<short>      , 0 /*pyrDown_gpu<short2>*/ , pyrDown_gpu<short3>      , pyrDown_gpu<short4>      },
        {0 /*pyrDown_gpu<int>*/  , 0 /*pyrDown_gpu<int2>*/   , 0 /*pyrDown_gpu<int3>*/  , 0 /*pyrDown_gpu<int4>*/  },
        {pyrDown_gpu<float>      , 0 /*pyrDown_gpu<float2>*/ , pyrDown_gpu<float3>      , pyrDown_gpu<float4>      }
    };

    GpuMat src = _src.getGpuMat();

    CV_Assert( src.depth() <= CV_32F && src.channels() <= 4 );

    const func_t func = funcs[src.depth()][src.channels() - 1];
    CV_Assert( func != 0 );

    _dst.create((src.rows + 1) / 2, (src.cols + 1) / 2, src.type());
    GpuMat dst = _dst.getGpuMat();

    func(src, dst, StreamAccessor::getStream(stream));
}

void cv::gpu::equalizeHist(InputArray _src, OutputArray _dst, InputOutputArray _buf, Stream& _stream)
{
    GpuMat src = _src.getGpuMat();

    CV_Assert( src.type() == CV_8UC1 );

    _dst.create(src.size(), src.type());
    GpuMat dst = _dst.getGpuMat();

    int intBufSize;
    nppSafeCall( nppsIntegralGetBufferSize_32s(256, &intBufSize) );

    size_t bufSize = intBufSize + 2 * 256 * sizeof(int);

    ensureSizeIsEnough(1, static_cast<int>(bufSize), CV_8UC1, _buf);
    GpuMat buf = _buf.getGpuMat();

    GpuMat hist(1, 256, CV_32SC1, buf.data);
    GpuMat lut(1, 256, CV_32SC1, buf.data + 256 * sizeof(int));
    GpuMat intBuf(1, intBufSize, CV_8UC1, buf.data + 2 * 256 * sizeof(int));

    gpu::calcHist(src, hist, _stream);

    cudaStream_t stream = StreamAccessor::getStream(_stream);
    NppStreamHandler h(stream);

    nppSafeCall( nppsIntegral_32s(hist.ptr<Npp32s>(), lut.ptr<Npp32s>(), 256, intBuf.ptr<Npp8u>()) );

    hist::equalizeHist(src, dst, lut.ptr<int>(), stream);
}

cv::GlBuffer::GlBuffer(InputArray mat_, Usage _usage) : rows_(0), cols_(0), type_(0), usage_(_usage)
{
#ifndef HAVE_OPENGL
    (void)mat_;
    (void)_usage;
    throw_nogl;
#else
    int kind = mat_.kind();
    Size _size = mat_.size();
    int _type = mat_.type();

    if (kind == _InputArray::GPU_MAT)
    {
        #if !defined HAVE_CUDA || defined(CUDA_DISABLER)
            throw_nocuda;
        #else
            GpuMat d_mat = mat_.getGpuMat();
            impl_ = new Impl(d_mat.rows, d_mat.cols, d_mat.type(), _usage);
            impl_->copyFrom(d_mat);
        #endif
    }
    else
    {
        Mat mat = mat_.getMat();
        impl_ = new Impl(mat, _usage);
    }

    rows_ = _size.height;
    cols_ = _size.width;
    type_ = _type;
#endif
}

void DepthmapDenoiseWeightedHuberImpl::cacheGValues(InputArray _visibleLightImage){
    using namespace cv::cuda::device::dtam_denoise;
    localStream = cv::cuda::StreamAccessor::getStream(cvStream);
    if (!_visibleLightImage.empty()){
        visibleLightImage=_visibleLightImage.getGpuMat();
        cachedG=0;
    }
    if(cachedG)
        return;//already cached
    if(!alloced)
        allocate(rows,cols);

    
    // Call the gpu function for caching g's
    
    loadConstants(rows, cols, 0, 0, 0, 0, 0, 0,
            0, 0);
    CV_Assert(_g1.isContinuous());
    float* pp = (float*) visibleLightImage.data;//TODO: write a color version.
    float* g1p = (float*)_g1.data;
    float* gxp = (float*)_gx.data;
    float* gyp = (float*)_gy.data;
    computeGCaller(pp,  g1p,  gxp,  gyp, cols);
    cachedG=1;
}

void cv::cuda::fastNlMeansDenoising(InputArray _src, OutputArray _dst, float h, int search_window, int block_window, Stream& stream)
{
    const GpuMat src = _src.getGpuMat();

    CV_Assert(src.depth() == CV_8U && src.channels() < 4);

    int border_size = search_window/2 + block_window/2;
    Size esize = src.size() + Size(border_size, border_size) * 2;

    BufferPool pool(stream);

    GpuMat extended_src = pool.getBuffer(esize, src.type());
    cv::cuda::copyMakeBorder(src, extended_src, border_size, border_size, border_size, border_size, cv::BORDER_DEFAULT, Scalar(), stream);
    GpuMat src_hdr = extended_src(Rect(Point2i(border_size, border_size), src.size()));

    int bcols, brows;
    device::imgproc::nln_fast_get_buffer_size(src_hdr, search_window, block_window, bcols, brows);
    GpuMat buffer = pool.getBuffer(brows, bcols, CV_32S);

    using namespace cv::cuda::device::imgproc;
    typedef void (*nlm_fast_t)(const PtrStepSzb&, PtrStepSzb, PtrStepi, int, int, float, cudaStream_t);
    static const nlm_fast_t funcs[] = { nlm_fast_gpu<uchar>, nlm_fast_gpu<uchar2>, nlm_fast_gpu<uchar3>, 0};

    _dst.create(src.size(), src.type());
    GpuMat dst = _dst.getGpuMat();

    funcs[src.channels()-1](src_hdr, dst, buffer, search_window, block_window, h, StreamAccessor::getStream(stream));
}

void cv::cuda::magnitudeSqr(InputArray _src, OutputArray _dst, Stream& stream)
{
    GpuMat src = _src.getGpuMat();

    _dst.create(src.size(), CV_32FC1);
    GpuMat dst = _dst.getGpuMat();

    npp_magnitude(src, dst, nppiMagnitudeSqr_32fc32f_C1R, StreamAccessor::getStream(stream));
}

GpuMat DepthmapDenoiseWeightedHuberImpl::operator()(InputArray _ain, float epsilon,float theta){
    const GpuMat& ain=_ain.getGpuMat();
    
    using namespace cv::cuda::device::dtam_denoise;
    localStream = cv::cuda::StreamAccessor::getStream(cvStream);
    
    rows=ain.rows;
    cols=ain.cols;
    
    CV_Assert(ain.cols>0);
    if(!(ain.rows % 32 == 0 && ain.cols % 32 == 0 && ain.cols >= 64)){
        CV_Assert(!"For performance reasons, DepthmapDenoiseWeightedHuber currenty only supports multiple of 32 image sizes with cols >= 64. Pad the image to achieve this.");
    }
    rows=ain.rows;
    cols=ain.cols;
    if(!ain.isContinuous()){
        _a.create(1,rows*cols, CV_32FC1);
        _a=_a.reshape(0,rows);
        ain.copyTo(_a,cvStream);
    }else{
        _a=ain;
    }
    

    
    if(!alloced){
        allocate(rows,cols);
    } 
    
    if(!visibleLightImage.empty())
        cacheGValues();
    if(!cachedG){
        _gx.setTo(1,cvStream);
        _gy.setTo(1,cvStream);
    }
    if(!dInited){
        _a.copyTo(_d,cvStream);
        dInited=1;
    }
    
    computeSigmas(epsilon,theta);
    
    float* d = (float*) _d.data;
    float* a = (float*) _a.data;
    float* gxpt = (float*)_gx.data;
    float* gypt = (float*)_gy.data;
    float* gqxpt = (float*)_qx.data;
    float* gqypt = (float*)_qy.data;

    loadConstants(rows, cols, 0, 0, 0, 0, 0, 0,
           0, 0);
    updateQDCaller  ( gqxpt, gqypt, d, a,
            gxpt, gypt, cols, sigma_q, sigma_d, epsilon, theta);
    cudaSafeCall(cudaGetLastError());
    return _d;
}

void cv::cuda::remap(InputArray _src, OutputArray _dst, InputArray _xmap, InputArray _ymap, int interpolation, int borderMode, Scalar borderValue, Stream& stream)
{
    using namespace cv::cuda::device::imgproc;

    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, PtrStepSzf xmap, PtrStepSzf ymap, PtrStepSzb dst, int interpolation,
        int borderMode, const float* borderValue, cudaStream_t stream, bool cc20);
    static const func_t funcs[6][4] =
    {
        {remap_gpu<uchar>      , 0 /*remap_gpu<uchar2>*/ , remap_gpu<uchar3>     , remap_gpu<uchar4>     },
        {0 /*remap_gpu<schar>*/, 0 /*remap_gpu<char2>*/  , 0 /*remap_gpu<char3>*/, 0 /*remap_gpu<char4>*/},
        {remap_gpu<ushort>     , 0 /*remap_gpu<ushort2>*/, remap_gpu<ushort3>    , remap_gpu<ushort4>    },
        {remap_gpu<short>      , 0 /*remap_gpu<short2>*/ , remap_gpu<short3>     , remap_gpu<short4>     },
        {0 /*remap_gpu<int>*/  , 0 /*remap_gpu<int2>*/   , 0 /*remap_gpu<int3>*/ , 0 /*remap_gpu<int4>*/ },
        {remap_gpu<float>      , 0 /*remap_gpu<float2>*/ , remap_gpu<float3>     , remap_gpu<float4>     }
    };

    GpuMat src = _src.getGpuMat();
    GpuMat xmap = _xmap.getGpuMat();
    GpuMat ymap = _ymap.getGpuMat();

    CV_Assert( src.depth() <= CV_32F && src.channels() <= 4 );
    CV_Assert( xmap.type() == CV_32F && ymap.type() == CV_32F && xmap.size() == ymap.size() );
    CV_Assert( interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC );
    CV_Assert( borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP );

    const func_t func = funcs[src.depth()][src.channels() - 1];
    if (!func)
        CV_Error(Error::StsUnsupportedFormat, "Unsupported input type");

    _dst.create(xmap.size(), src.type());
    GpuMat dst = _dst.getGpuMat();

    Scalar_<float> borderValueFloat;
    borderValueFloat = borderValue;

    Size wholeSize;
    Point ofs;
    src.locateROI(wholeSize, ofs);

    func(src, PtrStepSzb(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, xmap, ymap,
        dst, interpolation, borderMode, borderValueFloat.val, StreamAccessor::getStream(stream), deviceSupports(FEATURE_SET_COMPUTE_20));
}