Converting RGB to Grayscale with Jetson-Utils

[broger33]

I have been trying to use the imageIO library and cudaConvertColor function to load an RGB image, convert it to grayscale, then save the image to a new file.

Below is my code:

#include <jetson-utils/cudaColorspace.h>
#include <jetson-utils/cudaMappedMemory.h>
#include <jetson-utils/imageIO.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>

int main(int argc, char *argv[]) {
    uchar4  *src=NULL;
    uint8_t *dst=NULL;
    int     width=0, height=0;

    if (argc != 3) {
        fprintf(stderr, "Usage: %s image_source image_dest\r\n", argv[0]);
        return EXIT_FAILURE;
    }

    if (!loadImage(argv[1], (void**)&src, &width, &height, IMAGE_RGBA8)) {
        fprintf(stderr, "Error loading image\r\n");
        return EXIT_FAILURE;
    }
    cudaAllocMapped(&dst, width*height*sizeof(uint8_t));
    cudaConvertColor(src, IMAGE_RGBA8, dst, IMAGE_GRAY8, width, height);

    if (!saveImage(argv[2], dst, width, height)) {
        fprintf(stderr, "Failed to save image to %s\r\n", argv[2]);
    }

    cudaFree(src);
    cudaFree(dst);
    return EXIT_SUCCESS;
}

When trying to build the program, I ran into the following error.

/usr/local/include/jetson-utils/imageFormat.inl(234): error: static assertion failed with “invalid image format type - supported types are uchar3, uchar4, float3, float4”
static_assert(__image_format_assert_false::value, “invalid image format type - supported types are uchar3, uchar4, float3, float4”);
^
detected during:
instantiation of “imageFormat imageFormatFromType() [with T=uint8_t]”

I looked into the file that was throwing the error, and I found the following:

template<> inline imageFormat imageFormatFromType<uchar3>()	{ return IMAGE_RGB8; }
template<> inline imageFormat imageFormatFromType<uchar4>()	{ return IMAGE_RGBA8; }
template<> inline imageFormat imageFormatFromType<float3>()	{ return IMAGE_RGB32F; }
template<> inline imageFormat imageFormatFromType<float4>()	{ return IMAGE_RGBA32F; }

As an experiment, I added the following line, rebuilt, and the error went away. Doing this, I was able to perform the RGB to grayscale conversion with cudaConvertColor.

template<> inline imageFormat imageFormatFromType<uint8_t>() { return IMAGE_GRAY8; }

Is adding this line necessary? Are there other ways to setup the cudaConvertColor to perform RGB to grayscale that doesn’t involve modifying the underlying libraries/includes?

[kayccc]

Sorry for the late response.
Is this still an issue to support? Any result can be shared?

[broger33]

No worries, I was wondering if using cudaConvertColor to convert an RGB image to a grayscale image and save the grayscale image was supported.

I was getting a “static assert” error with the saveImage command when trying to save the grayscale image, but I was able to fix the error by adding an imageFormatFromType for grayscale types (i.e. uint8_t) as mentioned in the original post. As this modifies the underlying library code, specifically jetson-utils/imageFormat.inl, I was wondering if this is the best solution, or if there is a way to convert RGB to grayscale and save the grayscale image that doesn’t result in the static assert error.

[AastaLLL]

Hi

Thanks for sharing this info.

Is the output expected?
If yes, could you submit a PR to the jetson_utils GitHub so our internal can review and integrate it?

https://github.com/dusty-nv/jetson-utils/pulls

Thanks.