Source code dead link

[mjstevens777]

It looks like the original source code download at http://vision.csd.uwo.ca/code/gco-v3.0.zip has moved. I'm not sure if this is a temporary or a permanent issue.

[Borda]

It seems that all Vision lab's page is down...
just in case, the code is also here - https://github.com/Borda/pyGCO/tree/master/gco_source

[amueller]

@Borda do you have written permission by Olga Veksler to redistribute the code or are you breaking the license?

[amueller]

It explicitly says "Only Olga Veksler has the right to redistribute this code, unless expressed permission is given otherwise"

[Borda]

@amueller in such case, I do not have anything like this... :( so could you ask her or shall I try to write her?

[amueller]

I was not able to obtain anything, which is why I download it on demand from the website, which is the only legal thing do to (and also possibly kinda sketchy).

[amueller]

The copyright is actually with the university, I think, and she can't simply change it.

[Borda]

I recall that one contributor made a small change in the C code so that is the reason why we have it in the repository...

[amueller]

That doesn't make it legal though... Sent from phone. Please excuse spelling and brevity.

…

On Mon, Sep 3, 2018, 02:39 Jirka Borovec ***@***.***> wrote: I recall that one contributor made a small change in the C code so that is the reason why we have it in the repository... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#18 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAbcFq9LzcJvT1GE7hU49vU4ionS0regks5uXM6hgaJpZM4WJw3N> .

[amueller]

Have you looked at opengm? I'm pretty sure they have their own implementation. Opencv does, too Sent from phone. Please excuse spelling and brevity.

…

On Tue, Sep 4, 2018, 09:29 Andreas Mueller ***@***.***> wrote: That doesn't make it legal though... Sent from phone. Please excuse spelling and brevity. On Mon, Sep 3, 2018, 02:39 Jirka Borovec ***@***.***> wrote: > I recall that one contributor made a small change in the C code so that > is the reason why we have it in the repository... > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > <#18 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AAbcFq9LzcJvT1GE7hU49vU4ionS0regks5uXM6hgaJpZM4WJw3N> > . >

[kunal0895]

http://mouse.cs.uwaterloo.ca/code/gco-v3.0.zip

For this zip file, this link worked. Using pip for installing this fails since it refers to the previous link and returns a 404.

[NarayanSchuetz]

It's dead again ;)!

That's why conferences and journals should make it mandatory that code not only be available but be available in a public repository and not some University site that may or may not exist in the next few years to come.

[amueller]

I recommend using the OpenGM implementation - at least that was a good idea last time I checked quite a while ago.

[amueller]

Well there's a listing here, but it links to the dead link: https://vision.cs.uwaterloo.ca/code/

[Borda]

I recommend using the OpenGM implementation - at least that was a good idea last time I checked quite a while ago.

do you have an example of a suitable replacement, which function from OpenGM would you recommend?

EDIT: the problem could be similar as they claimed [UNMAINTAINED] OpenGM 2

[NarayanSchuetz]

I don't think OpenGM implements their extended Alpha-Extension / A-B-Swap with label costs.
I mean technically you are allowed to use their source code under their conditions but not allowed to redistribute it.

In the end it's very unlikely that anything would happen as the work most likely was created under employment by the University of Waterloo and thus belongs to them and was meant for non-commercial public access in the first place.

For custom implementations based on min-cuts I had success with this python wrapper based on Vladimir Kolmogorov's C++ source.

[amueller]

The alpha-extension and a-b-swap are relatively simple to implement based if you have the graphcuts. Maybe OpenCV would be a suitable implementation?

I also have PyQPBO which is more general (allows non-submodular potentials) but basically as fast as GCO:
https://github.com/pystruct/pyqpbo

It also has an implementation of alpha expansion based on QPBO that's not in the original implementation:
https://github.com/pystruct/pyqpbo/blob/master/src/pyqpbo.pyx#L267

[amueller]

fixed in #23.