alemart / speedy-vision

GPU-accelerated Computer Vision for JavaScript.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

feature matching

under-score opened this issue · comments

great to have now also LSH/ KNN feature matching.

Unfortunately - there are null matches also in the online demo https://alemart.github.io/speedy-vision/demos/feature-matching.html

Would you expand on the issue?

Did you find an issue with the demo?

Yes - see the screenshot of the demo - no error in the console, but always 0 matches.

Bildschirmfoto 2022-06-01 um 15 29 43

Does it work on a different browser/machine?
Does it work if you clear the cache of this browser?
What is your browser / OS?

Turned out that It is running on a newer Mac but have also an older 1,7 GHz Air here with Dual-Core Intel i7 & Big Sur 11.6.
where Chrome (but also FF and Safari) always evaluates keypoint.matches[0].index to -1

If the demo didn't work yesterday but is working now, the issue was probably related to an old version of the library being cached by the browser.

When you mention that keypoint.matches[0].index is evaluated to -1, are you talking about the demo as is or do you have a different code? Have you inspected the value?

sorry for delay - just tried it again - it is not a cache issue on the old machine.

I was talking about the demo, and yes, the value is always -1 also in the console.

Thank you for your reply.

I just submitted a slight change to a shader. Please let me know if you see any match at all (clear the cache first).

In addition, please try the demo in an operating system other than macOS if possible.

no issue with other machines and other OS's while the error still persists on the machine described above.

Unfortunately I'm unable to reproduce the issue in my machines.

If no matches are found, but the demo doesn't crash, then we have an indication that some shader in the pipeline isn't providing the expected output. Since this issue is happening on that specific machine, but not on the others, then there is the possibility of a bug in the video drivers - but I can't say for sure.

A deeper inspection of the output of the shaders in that specific machine could reveal the spot of failure and a possible workaround. However, I'm unable to do it by myself because my machines do not exhibit that failure.

thanks for all - I also has the impression that this is hardware dependent.