Python and Matlab bindings for Damascene, Global Probabilty of Boundary on GPU

So I'm still playing around with constrained parametric min-cuts for object segmentation. A major bottleneck of this algorithm is gPb, the global probability of boundary operator from Malik's group.
Luckily, there already is a CUDA implementation of gPb out there: Damascene.
Damascene provides a command line interface to apply gPb to ppm images.
Since I wanted to include it directly with cpmc, I wrote some mex-wrappers for Damascene. And since I would love to see more algorithms done in Python instead of Matlab, I wrote some Python wrappers, too.
You can find both, together with the current version of damascene here.
You need a working CUDA setup and the acm library to compile it.
Set your paths in common.mk and just "make" it.
For the matlab wrappers, you also need to adjust the matlab path in "bindings/Makefile" and "make gpb_mex" in that directory.
For the python wrappers, you have to compile damascene with "make shared=1". Then you can simply call gpu_gpb.py inside the bindings directory.

If you only have the students' version of Matlab, like me, you can use the file "common_32.mk" instead of common.mk to cross-compile damascene. You will probably have to adjust some paths in there, too.

I'd like to thank Bryan Catanzaro for making Damascene public. It's really a great tool!

Comments

  1. helboiJune 19, 2011 at 3:57 PM

    Hi,
    Thanks a lot for the useful work and sharing it with the community
    Before going to matlab wrappers, i just wanted to run damascene in C++. After successful compilation, I've got this runtime error:
    "Assertion failed: screenedEigVals.size() >= p_nEigNum"
    Do you happen having faced this error?
    Could you help me solve this problem?
    Thanks in advance
    Best regards

    ReplyDelete
  2. Andreas MuellerJune 21, 2011 at 11:14 AM

    Hi.
    Nope, I never ran into this problem. What command line arguments did you use? Did you try the default parameters?
    Cheers,
    Andy

    ReplyDelete
  3. helboiJune 21, 2011 at 4:09 PM

    I did not find any default parameters nor a sample ppm image. I just used some images of mine and played with the parameters values but still having the same error.
    Could you please send me an image and corresponding parameters that work
    Thanks

    ReplyDelete
  4. Andreas MuellerJune 21, 2011 at 4:27 PM

    There is a file called "polynesia.ppm" in the "damascene" subfolder.
    Just call damascene with this file and no other arguments
    like this (my damascene executable is in some release folder at that point)
    "./damascene ../../../damascene/polynesia.ppm "

    ReplyDelete
  5. helboiJune 21, 2011 at 4:36 PM

    Thanks for the hint. But this is what i've got

    Using cuda device 0: Quadro FX 3800
    Processing: polynesia.ppm, output in polynesiaPb.pgm and polynesia.pb

    Eig 9 Tol 0.001000 Texton 1Image found: 321 x 481 pixels
    Available 874888704 bytes on GPU
    >+< rgbUtoGrayF | 0.088000 | ms
    Convolving
    Beginning kmeans
    Changes: 150510
    Changes: 75038
    Changes: 49962
    Changes: 36991
    Changes: 30317
    Changes: 24616
    Changes: 20667
    Changes: 17589
    Changes: 15044
    Changes: 12782
    Changes: 10673
    Changes: 8873
    Changes: 7254
    Changes: 5789
    Changes: 4663
    15 iterations until termination
    Kmeans completed
    >+< texton | 235.102005 | ms
    >+< rgbUtoLab3F | 0.181000 | ms
    >+< normalizeLab | 0.004000 | ms
    >+< mirrorImage | 1.537000 | ms
    Beginning Local cues computation
    >+< Bgsmooth: | 45.691002 | ms
    >+< Bg: | 276.975006 | ms
    >+< Cgsmooth: | 101.719002 | ms
    >+< Cga: | 315.234985 | ms
    >+< Cgsmooth: | 101.391998 | ms
    >+< Cgb: | 317.585999 | ms
    >+< Tgsmooth: | 97.671997 | ms
    >+< Tg: | 254.839996 | ms
    Completed Local cues
    localcues time: 1.198925 seconds
    >+< localcues | 1198.927979 | ms
    >+< combine | 0.374000 | ms

    Max time: 0.000117 seconds
    Oriented Max time: 0.001017 seconds
    Solve time: 0.001145 seconds
    >+< nonmaxsupression | 3.273000 | ms
    Intervening contour completed
    >+< intervene | 37.535000 | ms
    Available 804051968 bytes on GPU
    Can fit 1158 iterations on GPU
    lanczos iteration: 0
    lanczos iteration: 100
    lanczos iteration: 200
    lanczos iteration: 300
    lanczos iteration: 400
    Screened Eigenvalues:
    nan nan 0.000000e+00
    damascene: lanczos.cu:360: void CullumDevice(int, float*, float*, double*, double*, int, float*, float*, int, float*, double*, char, char, double, double, int, int, double, int, double*, int*, int*, double*, int*, int*): Assertion `screenedEigVals.size() >= p_nEigNum' failed.
    Abandon

    ReplyDelete
  6. Andreas MuellerJune 21, 2011 at 4:48 PM

    Weird.
    Seems to be some numerical problem. Which version of damascene are you using? I'm using 0.1.4. Can you try another GPU?
    Maybe try the CUDA memtest tool: http://sourceforge.net/projects/cudagpumemtest/

    ReplyDelete
  7. helboiJune 21, 2011 at 6:38 PM

    I used the source files provided by your package (0.1.4) and finally it works! :))
    It's weird that in the damascene page (http://www.cs.berkeley.edu/~catanzar/Damascene/),the version provided is 0.1 and not the 0.1.4.
    I wonder where have you downloaded the 0.1.4 version?
    Thanks a lot anyway for helping me :)

    ReplyDelete
  8. Andreas MuellerJune 21, 2011 at 6:47 PM

    There is another place to download it. Follow the link at "favourite code" at the right.
    I notice they have 0.1.5 now.
    Since you had so much trouble with the older version, maybe you should tell them to update their links.
    Cheers,
    Andy

    ReplyDelete
  9. Andreas MuellerJune 21, 2011 at 6:58 PM

    There was a minor fix in 0.1.5, which I included in my matlab and python bindings. If you want, you can get them here:
    https://github.com/amueller/damascene-python-and-matlab-bindings

    ReplyDelete
  10. helboiJune 21, 2011 at 7:41 PM

    I just sent an e-mail to Damascene author

    ReplyDelete

Post a Comment

Popular posts from this blog

Machine Learning Cheat Sheet (for scikit-learn)

Image
As you hopefully have heard, we at scikit-learn are doing a user survey (which is still open by the way). One of the requests there was to provide some sort of flow chart on how to do machine learning. As this is clearly impossible, I went to work straight away. This is the result: [edit2] clarification: With ensemble classifiers and ensemble regressors I mean random forests , extremely randomized trees, gradient boosted trees , and the soon-to-be-come weight boosted trees (adaboost). [/edit2] Needless to say, this sheet is completely authoritative.
Read more

A Wordcloud in Python

Image
Last week I was at Pycon DE , the German Python conference. After hacking on scikit-learn a lot last week, I decided to to something different on my way back, that I had planned for quite a while: doing a wordl -like word cloud . I know, word clouds are a bit out of style but I kind of like them any way. My motivation to think about word clouds was that I thought these could be combined with topic-models to give somewhat more interesting visualizations. So I looked around to find a nice open-source implementation of word-clouds ... only to find none. (This has been a while, maybe it has changed since). While I was bored in the train last week, I came up with this code . A little today-themed taste:
Read more

MNIST for ever....

Image
[update] This post is a bit old, but many people still seem interested. So just a short update: Nowadays I would use Python and scikit-learn to do this. Here is an example of how to do cross-validation for SVMs in scikit-learn. Scikit-learn even downloads MNIST for you. [/update] MNIST is, for better or worse, one of the standard benchmarks for machine learning and is also widely used in then neural networks community as a toy vision problem. Just for the unlikely case that anyone is not familiar with it: It is a dataset of handwritten digits, 0-9, in black on white background. It looks something like this: There are 60000 training and 10000 test images, each 28x28 gray scale. There are roughly the same number of examples of each category in the test and training datasets. I used it in some papers myself even though there are some reasons why it is a little weird. Some not-so-obvious (or maybe they are) facts are: - The images actually contain a 20x20 patch of digi...
Read more

Python things you never need: Empty lambda functions

Ever wanted to create an empty Python function in-line? No? Me neither. But my coworker does... We found the answer: <pre> x = lambda : None </pre> If you ever needed this, please tell me ;) [Edit] Apparently some people do need it :) Btw, most people seem to try Most people try lambda: pass first. But you have to remember, the thing at the right side of the colon is the return value of the function, not the body! So an expression is needed, not a statement. [\Edit]
Read more