Jetson Inference
DNN Vision Library

Colorspace conversion functions for various YUV formats, RGB, BGR, Bayer, and grayscale. More...

Functions

cudaError_t cudaConvertColor (void *input, imageFormat inputFormat, void *output, imageFormat outputFormat, size_t width, size_t height, const float2 &pixel_range=make_float2(0, 255))
 Convert between two image formats using the GPU. More...
 
template<typename T_in , typename T_out >
cudaError_t cudaConvertColor (T_in *input, T_out *output, size_t width, size_t height, const float2 &pixel_range=make_float2(0, 255))
 Convert between to image formats using the GPU. More...
 

8-bit grayscale to floating-point grayscale (and vice versa)

See also
cudaConvertColor() from cudaColorspace.h for automated format conversion
cudaError_t cudaGray8ToGray32 (uint8_t *input, float *output, size_t width, size_t height)
 Convert uint8 grayscale image into float grayscale. More...
 
cudaError_t cudaGray32ToGray8 (float *input, uint8_t *output, size_t width, size_t height, const float2 &pixelRange=make_float2(0, 255))
 Convert float grayscale image into uint8 grayscale. More...
 

RGB/BGR to 8-bit grayscale

See also
cudaConvertColor() from cudaColorspace.h for automated format conversion
cudaError_t cudaRGB8ToGray8 (uchar3 *input, uint8_t *output, size_t width, size_t height, bool swapRedBlue=false)
 Convert uchar3 RGB/BGR image into uint8 grayscale. More...
 
cudaError_t cudaRGBA8ToGray8 (uchar4 *input, uint8_t *output, size_t width, size_t height, bool swapRedBlue=false)
 Convert uchar4 RGBA/BGRA image into uint8 grayscale. More...
 
cudaError_t cudaRGB32ToGray8 (float3 *input, uint8_t *output, size_t width, size_t height, bool swapRedBlue=false, const float2 &pixelRange=make_float2(0, 255))
 Convert float3 RGB/BGR image into uint8 grayscale. More...
 
cudaError_t cudaRGBA32ToGray8 (float4 *input, uint8_t *output, size_t width, size_t height, bool swapRedBlue=false, const float2 &pixelRange=make_float2(0, 255))
 Convert float4 RGBA/BGRA image into uint8 grayscale. More...
 

RGB/BGR to floating-point grayscale

See also
cudaConvertColor() from cudaColorspace.h for automated format conversion
cudaError_t cudaRGB8ToGray32 (uchar3 *input, float *output, size_t width, size_t height, bool swapRedBlue=false)
 Convert uchar3 RGB/BGR image into float grayscale. More...
 
cudaError_t cudaRGBA8ToGray32 (uchar4 *input, float *output, size_t width, size_t height, bool swapRedBlue=false)
 Convert uchar4 RGBA/BGRA image into float grayscale. More...
 
cudaError_t cudaRGB32ToGray32 (float3 *input, float *output, size_t width, size_t height, bool swapRedBlue=false)
 Convert float3 RGB/BGR image into float grayscale. More...
 
cudaError_t cudaRGBA32ToGray32 (float4 *input, float *output, size_t width, size_t height, bool swapRedBlue=false)
 Convert float4 RGB/BGR image into float grayscale. More...
 

8-bit grayscale to RGB/BGR

See also
cudaConvertColor() from cudaColorspace.h for automated format conversion
cudaError_t cudaGray8ToRGB8 (uint8_t *input, uchar3 *output, size_t width, size_t height)
 Convert uint8 grayscale image into uchar3 RGB/BGR. More...
 
cudaError_t cudaGray8ToRGBA8 (uint8_t *input, uchar4 *output, size_t width, size_t height)
 Convert uint8 grayscale image into uchar4 RGB/BGR. More...
 
cudaError_t cudaGray8ToRGB32 (uint8_t *input, float3 *output, size_t width, size_t height)
 Convert uint8 grayscale image into float3 RGB/BGR. More...
 
cudaError_t cudaGray8ToRGBA32 (uint8_t *input, float4 *output, size_t width, size_t height)
 Convert uint8 grayscale image into float4 RGB/BGR. More...
 

Floating-point grayscale to RGB/BGR

See also
cudaConvertColor() from cudaColorspace.h for automated format conversion
cudaError_t cudaGray32ToRGB8 (float *input, uchar3 *output, size_t width, size_t height, const float2 &pixelRange=make_float2(0, 255))
 Convert float grayscale image into uchar3 RGB/BGR. More...
 
cudaError_t cudaGray32ToRGBA8 (float *input, uchar4 *output, size_t width, size_t height, const float2 &pixelRange=make_float2(0, 255))
 Convert float grayscale image into uchar4 RGB/BGR. More...
 
cudaError_t cudaGray32ToRGB32 (float *input, float3 *output, size_t width, size_t height)
 Convert float grayscale image into float3 RGB/BGR. More...
 
cudaError_t cudaGray32ToRGBA32 (float *input, float4 *output, size_t width, size_t height)
 Convert float grayscale image into float4 RGB/BGR. More...
 

Detailed Description

Colorspace conversion functions for various YUV formats, RGB, BGR, Bayer, and grayscale.

See also
cudaConvertColor for automated format conversion.
cudaYUV.h for the YUV functions.

Function Documentation

◆ cudaConvertColor() [1/2]

cudaError_t cudaConvertColor ( void *  input,
imageFormat  inputFormat,
void *  output,
imageFormat  outputFormat,
size_t  width,
size_t  height,
const float2 &  pixel_range = make_float2(0, 255) 
)

Convert between two image formats using the GPU.

This function supports various conversions between RGB/RGBA, BGR/BGRA, Bayer, grayscale, and YUV. In addition to converting between the previously listed colorspaces, you can also change the number of channels or data format (for example, IMAGE_RGB8 to IMAGE_RGBA32F).

For the list of image formats available,

See also
imageFormat enum.

Limitations and unsupported conversions include:

- The YUV formats don't support BGR/BGRA or grayscale (RGB/RGBA only)
- YUV NV12, YUYV, YVYU, and UYVY can only be converted to RGB/RGBA (not from)
- Bayer formats can only be converted to RGB8 (`uchar3`) and RGBA8 (`uchar4`)
Parameters
inputCUDA device pointer to the input image
inputFormatformat enum of the input image
outputCUDA device pointer to the input image
outputFormatformat enum of the output image
widthwidth of the input and output images (in pixels)
heightheight of the input and output images (in pixels)
pixel_rangefor floating-point to 8-bit conversions, specifies the range of pixel intensities in the input image that get normalized to [0,255]. The default input range is [0,255], and as such no normalization occurs. Other common pixel ranges include [0,1] and [-1,1], and these pixel values would be re-scaled for [0,255] output. Note that this parameter is only used for float-to-uchar conversions where the data is downcast (for example, IMAGE_RGB32F to IMAGE_RGB8).

◆ cudaConvertColor() [2/2]

template<typename T_in , typename T_out >
cudaError_t cudaConvertColor ( T_in *  input,
T_out *  output,
size_t  width,
size_t  height,
const float2 &  pixel_range = make_float2(0,255) 
)

Convert between to image formats using the GPU.

This templated overload of cudaConvertColor() supports uchar3 (IMAGE_RGB8), uchar4 (IMAGE_RGBA8), float3 (IMAGE_RGB32F), and float4 (IMAGE_RGBA32F).

To convert to/from other formats such as YUV, grayscale, and BGR/BGRA, see the other version of cudaConvertColor() that uses explicity imageFormat enums.

Parameters
inputCUDA device pointer to the input image
outputCUDA device pointer to the input image
widthwidth of the input and output images (in pixels)
heightheight of the input and output images (in pixels)
pixel_rangefor floating-point to 8-bit conversions, specifies the range of pixel intensities in the input image that get normalized to [0,255]. The default input range is [0,255], and as such no normalization occurs. Other common pixel ranges include [0,1] and [-1,1], and these pixel values would be re-scaled for [0,255] output. Note that this parameter is only used for float-to-uchar conversions where the data is downcast (for example, IMAGE_RGB32F to IMAGE_RGB8).

◆ cudaGray32ToGray8()

cudaError_t cudaGray32ToGray8 ( float *  input,
uint8_t *  output,
size_t  width,
size_t  height,
const float2 &  pixelRange = make_float2(0, 255) 
)

Convert float grayscale image into uint8 grayscale.

Parameters
pixelRangespecifies the floating-point pixel value range of the input image, which is used to rescale the fixed-point pixel outputs to [0,255]. The default input range is [0,255], where no rescaling occurs. Other common input ranges are [-1, 1] or [0,1].

◆ cudaGray32ToRGB32()

cudaError_t cudaGray32ToRGB32 ( float *  input,
float3 *  output,
size_t  width,
size_t  height 
)

Convert float grayscale image into float3 RGB/BGR.

◆ cudaGray32ToRGB8()

cudaError_t cudaGray32ToRGB8 ( float *  input,
uchar3 *  output,
size_t  width,
size_t  height,
const float2 &  pixelRange = make_float2(0, 255) 
)

Convert float grayscale image into uchar3 RGB/BGR.

Parameters
pixelRangespecifies the floating-point pixel value range of the input image,
pixelRangespecifies the floating-point pixel value range of the input image, which is used to rescale the fixed-point pixel outputs to [0,255]. The default input range is [0,255], where no rescaling occurs. Other common input ranges are [-1, 1] or [0,1].

◆ cudaGray32ToRGBA32()

cudaError_t cudaGray32ToRGBA32 ( float *  input,
float4 *  output,
size_t  width,
size_t  height 
)

Convert float grayscale image into float4 RGB/BGR.

◆ cudaGray32ToRGBA8()

cudaError_t cudaGray32ToRGBA8 ( float *  input,
uchar4 *  output,
size_t  width,
size_t  height,
const float2 &  pixelRange = make_float2(0, 255) 
)

Convert float grayscale image into uchar4 RGB/BGR.

Parameters
pixelRangespecifies the floating-point pixel value range of the input image, which is used to rescale the fixed-point pixel outputs to [0,255]. The default input range is [0,255], where no rescaling occurs. Other common input ranges are [-1, 1] or [0,1].

◆ cudaGray8ToGray32()

cudaError_t cudaGray8ToGray32 ( uint8_t *  input,
float *  output,
size_t  width,
size_t  height 
)

Convert uint8 grayscale image into float grayscale.

◆ cudaGray8ToRGB32()

cudaError_t cudaGray8ToRGB32 ( uint8_t *  input,
float3 *  output,
size_t  width,
size_t  height 
)

Convert uint8 grayscale image into float3 RGB/BGR.

◆ cudaGray8ToRGB8()

cudaError_t cudaGray8ToRGB8 ( uint8_t *  input,
uchar3 *  output,
size_t  width,
size_t  height 
)

Convert uint8 grayscale image into uchar3 RGB/BGR.

◆ cudaGray8ToRGBA32()

cudaError_t cudaGray8ToRGBA32 ( uint8_t *  input,
float4 *  output,
size_t  width,
size_t  height 
)

Convert uint8 grayscale image into float4 RGB/BGR.

◆ cudaGray8ToRGBA8()

cudaError_t cudaGray8ToRGBA8 ( uint8_t *  input,
uchar4 *  output,
size_t  width,
size_t  height 
)

Convert uint8 grayscale image into uchar4 RGB/BGR.

◆ cudaRGB32ToGray32()

cudaError_t cudaRGB32ToGray32 ( float3 *  input,
float *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false 
)

Convert float3 RGB/BGR image into float grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.

◆ cudaRGB32ToGray8()

cudaError_t cudaRGB32ToGray8 ( float3 *  input,
uint8_t *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false,
const float2 &  pixelRange = make_float2(0, 255) 
)

Convert float3 RGB/BGR image into uint8 grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.
pixelRangespecifies the floating-point pixel value range of the input image, which is used to rescale the fixed-point pixel outputs to [0,255]. The default input range is [0,255], where no rescaling occurs. Other common input ranges are [-1, 1] or [0,1].

◆ cudaRGB8ToGray32()

cudaError_t cudaRGB8ToGray32 ( uchar3 *  input,
float *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false 
)

Convert uchar3 RGB/BGR image into float grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.

◆ cudaRGB8ToGray8()

cudaError_t cudaRGB8ToGray8 ( uchar3 *  input,
uint8_t *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false 
)

Convert uchar3 RGB/BGR image into uint8 grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.

◆ cudaRGBA32ToGray32()

cudaError_t cudaRGBA32ToGray32 ( float4 *  input,
float *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false 
)

Convert float4 RGB/BGR image into float grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.

◆ cudaRGBA32ToGray8()

cudaError_t cudaRGBA32ToGray8 ( float4 *  input,
uint8_t *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false,
const float2 &  pixelRange = make_float2(0, 255) 
)

Convert float4 RGBA/BGRA image into uint8 grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.
pixelRangespecifies the floating-point pixel value range of the input image, which is used to rescale the fixed-point pixel outputs to [0,255]. The default input range is [0,255], where no rescaling occurs. Other common input ranges are [-1, 1] or [0,1].

◆ cudaRGBA8ToGray32()

cudaError_t cudaRGBA8ToGray32 ( uchar4 *  input,
float *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false 
)

Convert uchar4 RGBA/BGRA image into float grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.

◆ cudaRGBA8ToGray8()

cudaError_t cudaRGBA8ToGray8 ( uchar4 *  input,
uint8_t *  output,
size_t  width,
size_t  height,
bool  swapRedBlue = false 
)

Convert uchar4 RGBA/BGRA image into uint8 grayscale.

Parameters
swapRedBlueif true, swap the input's red and blue channels. The default is false, and the channels will remain the same.