VPI 1.1.12 VPI_ERROR_INVALID_IMAGE_FORMAT: Input format's color range must be same as output'sa

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hi there,

I’m trying to convert VPIImage from VPI_IMAGE_FORMAT_UYVY to VPI_IMAGE_FORMAT_BGRA8 using the following code(not a complete code):

VPIStream stream;
CHECK_VPI_STATUS(vpiStreamCreate(0, &stream));

int w, h;
// img_data is the source data from v4l2 captured, image format is V4L2_PIX_FMT_UYVY (UYVY 4:2:2)
// img_frame is the source cv::Mat stored img_data
// img is the source image with format VPI_IMAGE_FORMAT_UYVY
unsigned char *img_data;
img_frame = cv::Mat(img_size, CV_8UC2, img_data);
CHECK_STATUS(vpiImageCreateOpenCVMatWrapper(img_frame, VPI_IMAGE_FORMAT_UYVY, 0, &img));
CHECK_VPI_STATUS(vpiImageGetSize(img, &w, &h));

VPIImage output;
CHECK_VPI_STATUS(vpiImageCreate(w, h, VPI_IMAGE_FORMAT_BGRA8, 0, &output));
CHECK_VPI_STATUS(vpiSubmitConvertImageFormat(stream, VPI_BACKEND_VIC, img, output, 0));
CHECK_VPI_STATUS(vpiStreamSync(stream));

According to documentation: VPI - Vision Programming Interface: Convert Image Format(VPI - Vision Programming Interface: Convert Image Format), this conversion is supported in VIC backend given scale == 1 and offset == 0, which is my case. I’m getting the following error:

VPI_ERROR_INVALID_IMAGE_FORMAT: Input format’s color range must be same as output’s

UYVY and BGRA8 have different color ranges, but the operation is listed as supported. Am I missing something?

Some more details:

hardware: AGX Xavier
Jetpack 4.6
VPI library version 1.1.12

And then:

My purpose is to convert UYVY to BGR8, if I operate as follows:
UYVY → UYVY_ER (by VIC)
UYVY_ER → BGRA8 (by VIC)
BGRA8 → BGR8 (by CUDA)

// use follows to get the final BGR8 data img_rgb_frame_vpi, BGR8 here represents the BGR8 VPIImage
VPIImageData out_data;
cv::Mat img_rgb_frame_from_vpi;
CHECK_STATUS(vpiImageLock(BGR8, VPI_LOCK_READ, &out_data));
CHECK_STATUS(vpiImageDataExportOpenCVMat(out_data, img_rgb_frame_vpi));
CHECK_STATUS(vpiImageUnlock(BGR8));

It works, but another question appears:

when using opencv cvtColor to get BGR8 from the same source img data like this:
cv::cvtColor(img_frame, img_rgb_frame_from_cv, cv::COLOR_YUV2BGR_UYVY);

cv::imwrite(“UYVY_TO_BGR8_BY_VPI.png”, img_rgb_frame_from_vpi);
cv::imwrite(“UYVY_TO_BGR8_BY_CV.png”, img_rgb_frame_from_cv);

I think that the “UYVY_TO_BGR8_BY_VPI.png” and “UYVY_TO_BGR8_BY_CV.png” should be almost the same.

In fact, I find that the difference between “UYVY_TO_BGR8_BY_VPI.png” and “UYVY_TO_BGR8_BY_CV.png” is distinct.

eg: The difference looks like this:

“UYVY_TO_BGR8_BY_CV.png”:
[51 59 55]
[52 60 57]
[53 61 58]

“UYVY_TO_BGR8_BY_VPI.png”:
[49 60 53]
[50 61 54]
[51 62 55]

So, my first question is whether the vpi convert can get the same result as opencv cvtColor,
secondly, by converting UYVY to UYVY_ER, then UYVY_ER to BGRA8, finally BGRA8 to BGR8, whether this way would cause more conversion errors,
lastly, is there any way to convert UYVY(YUV422) to BGR8 format by vpi directly?

2

Hello,

Looks like you posted in the wrong category, this is the Networking section. I will move this over to the Jetson AGX Xavier forum so the proper team can assist.

4

Hi,

This is a known issue.

Please use _ER format for VIC input.
The document should be updated. We will check this with our internal team.

For the comparison between OpenCV and VPI:
Since the hardware and implementation are different, we don’t guarantee that the result will be exactly the same as OpenCV.

But it should be visually close for the same input image.

Thanks.

6

Hi, @AastaLLL
Thanks for your reply, but I still have some doubts:
the input source image data comes from v4l2 captured,
and its image format is UYVY4:2:2, according to some information, its range might be same as studio range(as the document list: VPI - Vision Programming Interface: Convert Image Format),
so is that ok to use UYVY_ER to warp the souce data ?
eg:
CHECK_STATUS(vpiImageCreateOpenCVMatWrapper(img_frame, VPI_IMAGE_FORMAT_UYVY_ER, 0, &img));
CHECK_STATUS(vpiImageCreateOpenCVMatWrapper(img_frame, VPI_IMAGE_FORMAT_UYVY, 0, &img));
Are these two ways perform the same result?

7

Hi,

Do you have a chance to give it a try?
If it doesn’t work, you can convert the data from VPI_IMAGE_FORMAT_UYVY into VPI_IMAGE_FORMAT_UYVY_ER with a color convertor.

Thanks.

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.