Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

OpenCV 3.0.0 #5

Closed
jlaura opened this issue Jan 22, 2016 · 6 comments
Closed

OpenCV 3.0.0 #5

jlaura opened this issue Jan 22, 2016 · 6 comments

Comments

@jlaura
Copy link

jlaura commented Jan 22, 2016

First, thanks for serving these via binstar - immensely helpful to all of us. The upgrade to 3.1.0 causes breakage in the Python API for the Flann Matcher.

Seeing this, I forked your project and rolled back to the last commit before the 3.1.0 upgrade. Builds are successful via Travis (Linux) and a local OS X install (10.11.1). I am though seeing an intermittent memory related error using the Flann Matcher in 3.0.0 (specifically when adding descriptors). This is not something I saw in your original builds.

Is this something that you encountered and patched?

Is your 3.0.0 build replaced by the 3.1.0 build on binstar? I assume so, but figured I would ask.

@patricksnape
Copy link
Contributor

Oh darn!

I'm very sorry about that - this puts me in a bit of a bind. I honestly have just been building these for the sake of the community and I don't extensively use OpenCV outside of the face detection module. I also don't run an exhaustive test suite on these builds as I have been assuming that the tagged releases from OpenCV don't contain any breaking bugs - looks like I am wrong!

The biggest issue is that (unfortunately) my free account on binstar/anaconda only allows 2GB of file space and these OpenCV 3.x builds take ~1GB per version. So I had to delete the 3.0 build just to fit the 3.1 build on my account. Normally I would prefer to keep all versions I build, but I simply can't on the Anaconda free tier.

This commit should be the last commit that actually works - before I started messing around with Python 3.5 support. Is this commit still broken? Are you seeing the errors on both Linux and OSX?

@jlaura
Copy link
Author

jlaura commented Jan 22, 2016

I will give that commit a go and see if I can get a clean build. If so, I am glad to host openCV 3.0.0 on my binstar account and will serve for all 3 platforms. Hopefully the OpenCV team is able to address the 3.1.0 binding bug at somepoint in the near future (though it does not sound like a high priority yet).

@jlaura
Copy link
Author

jlaura commented Jan 22, 2016

@patricksnape Here are travis builds for both Linux and OS X. I did bump a few commits ahead to get Python 3.5 support. The files are available via binstar. I am still seeing the intermittent memory issue.

If you have a moment, does anything look out of sorts with those builds? I see a few warnings, but no errors.

@jlaura
Copy link
Author

jlaura commented Jan 22, 2016

I think I found the issue. The openCV3.0.0 release happened June 3. This PR, that looks like it will fix the issue was merged June 10. I am going to try altering the git clone in the .travis.yml to see if I can rollback to this PR.

@jlaura
Copy link
Author

jlaura commented Jan 22, 2016

Success. Thanks for the info! The 3.0.0 release with Python 3.5 support is now available via my binstar.
https://anaconda.org/jlaura/opencv3
https://travis-ci.org/USGS-Astrogeology/conda-opencv3

@jlaura jlaura closed this as completed Jan 22, 2016
@patricksnape
Copy link
Contributor

Great - glad to hear it! Though that bug does appear to be in 3.1 and I was also seeing segmentation faults on my OSX build even going back to 3.0 😕. I tried to git bisect but its super difficult to bisect and keep in sink both opencv_contrib and opencv. Since they need to be in sync but only the actual tagged releases have any guarantee of working together!

Unfortunately this FLANN matcher bug needs SIFT or SURF so contrib is required - which is frustrating.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants