Could not use "Stereo Disparity Estimator" with OFA backend

[dandelion1124]

Summary

Jetson Orin NX supports OFA backend in VPI.
So, I tried to use “Stereo Disparity Estimator” sample script with OFA backend.
But, I got the following error.

ValueError: VPI_ERROR_INVALID_IMAGE_FORMAT: Unsupported disparity format

Environment

• Jetson Orin NX(16GB)
• JetPack 5.1
• VPI 2.2.4
  • I used Python API

Steps to reproduce

$ cd /opt/nvidia/vpi2/samples/02-stereo_disparity
$ python3 main.py ofa ../assets/chair_stereo_left.png ../assets/chair_stereo_right.png
Traceback (most recent call last):
  File "main.py", line 123, in <module>
    disparity = vpi.stereodisp(left, right, downscale = downscale, out_confmap=confidenceMap, window=5, maxdisp=maxDisparity)
ValueError: VPI_ERROR_INVALID_IMAGE_FORMAT: Unsupported disparity format

[AastaLLL]

Hi,

Please use ofa-pva-vic instead.

Thanks.

[dandelion1124]

@AastaLLL Thank you for your reply.

Please use ofa-pva-vic instead.

It works fine.

This script has the following backend option for OFA backend.

• ofa
• ofa-pva-vic

But, the user needs to convert image format to use vpi.stereodisp with OFA backend.
And, this conversion supports only VIC backend.
https://docs.nvidia.com/vpi/2.2/algo_imageconv.html

Therefore, I think that ofa-vic is more appropriate than ofa.
Is my understanding correct?

And, I found that the behavior of backend option is different between C++ sample code and python sample code.
So, the C++ sample code works fine with ofa option.

[AastaLLL]

Hi,

Thanks for reporting this.

This looks like a bug in our python sample.
We are checking this with our internal team to fix this.

[AastaLLL]

Hi,

This issue is fixed in VPI 2.2.6.
You can find the package in the recently released JetPack5.1.1

Thanks.

[dandelion1124]

Thank you for your information! I’ll use the latest VPI.

[dandelion1124]

I checked that this problem was fixed in VPI 2.2.6.