[FeatureRequest] [VisionWorks] fidelity of Block Matching

I’m comparing the outputs of the Block Matching outputs of VisionWorks vs OpenCV’s calls, and have noticed that while the SGBM implementation outputs data in Q11.4 format, the simple BM does not. Whereas OpenCV outputs Q11.4 in both cases.

The result is a lot of banding in the stereo output, which quickly becomes very noticeable.

I’d like to see the Stereo Block Matching algorithm on VisionWorks provide increased fidelity, same as SGBM on VisionWorks and both implementations in OpenCV. Otherwise, really SGBM is the only useful implementation and on the TX2 can still be slow.

I would encourage you to provide this request via bug report at developer.nvidia.com

You can put the acronym RFE in the title, to help indicate it is a request for enhancement, not an actual defect.

Hello @txbob, I’m looking at developer.nvidia.com and I don’t see a bug tracker anywhere. Could you point me to the place where I can put in this RFE?

Thank you!

login as a registered developer
click on your name in the upper right hand corner
click on my account
click on my bugs


We have received your feature request and already report to the internal team.
Will update information with you once we got feedback.