Mailman 3 August 2012 - scikit-image

Re: Subpackage for utilities
by Stéfan van der Walt 20 Aug '12

20 Aug '12

Hi Tony On Thu, Aug 16, 2012 at 7:06 PM, Tony Yu <tsyu80(a)gmail.com> wrote: > In PR #239, Johannes suggested a new `utils` module or subpackage, which I > think this is a great idea. +1 > We currently have the `util` package, but I'm inclined to suggest a new > subpackage (not named `utils` to avoid confusion). How about ``skimage._algorithms`` or, based upon another suggestion on this thread, ``skimage._shared``? Stéfan

1 0

Re: Subpackage for utilities
by Andreas Mueller 18 Aug '12

18 Aug '12

Hi Tony. As I said, I think it is a great idea to collect the utility functions. I'm undecided on whether it is a good idea to separate them from the user utility functions. To the candidates I'd like to add the union find that I reused in my segmentation PR. Cheers, Andy On 08/17/2012 03:06 AM, Tony Yu wrote: > In PR #239 > <https://github.com/scikits-image/scikits-image/pull/239#issuecomment-7792726>, > Johannes suggested a new `utils` module or subpackage, which I think > this is a great idea. > > We currently have the `util` package, but I'm inclined to suggest a > new subpackage (not named `utils` to avoid confusion). It would be > nice to separate functions that would be useful to consumers of image > processing functions (e.g., the dtype conversion routines in `util`) > from functions that are useful to developers of image processing > functions (e.g., the `shape` module in the `util` package, which has > `view_as_blocks` and `view_as_windows`). I wouldn't be surprised if > people think this is an arbitrary (or false) distinction. > > Regardless of whether there should be a new subpackage (or if it > should be put in `util`), there are a number of candidate functions > that come to mind: > > - rank order (`filter` package) > - heap (`graph` package) > - heap_* (`morphology` package) (*.pxi files?) > - integral images (`transform` package). (The transform package does > make sense for this, but keep in mind that this is used in other > packages, e.g., `match_template` in the `feature` package.) > > There are probably a number of others that I'm missing. > > In the past, there have been discussions of dependency versions and > adding compatibility functions so that we can support older versions > of numpy, matplotlib, etc. A utility package would be good for those > sorts of things. > > -Tony

2 1

Subpackage for utilities
by Tony Yu 17 Aug '12

17 Aug '12

In PR #239<https://github.com/scikits-image/scikits-image/pull/239#issuecomment-7792726>, Johannes suggested a new `utils` module or subpackage, which I think this is a great idea. We currently have the `util` package, but I'm inclined to suggest a new subpackage (not named `utils` to avoid confusion). It would be nice to separate functions that would be useful to consumers of image processing functions (e.g., the dtype conversion routines in `util`) from functions that are useful to developers of image processing functions (e.g., the `shape` module in the `util` package, which has `view_as_blocks` and `view_as_windows`). I wouldn't be surprised if people think this is an arbitrary (or false) distinction. Regardless of whether there should be a new subpackage (or if it should be put in `util`), there are a number of candidate functions that come to mind: - rank order (`filter` package) - heap (`graph` package) - heap_* (`morphology` package) (*.pxi files?) - integral images (`transform` package). (The transform package does make sense for this, but keep in mind that this is used in other packages, e.g., `match_template` in the `feature` package.) There are probably a number of others that I'm missing. In the past, there have been discussions of dependency versions and adding compatibility functions so that we can support older versions of numpy, matplotlib, etc. A utility package would be good for those sorts of things. -Tony

1 0

Re: EuroSciPy sprint
by Andreas Müller 16 Aug '12

16 Aug '12

Hey Everybody. Just as a side node: If we had something like BRIEF, we could start with some serious vision applications. With key point matching we can do stitching, getting camera transforms etc. Using dense features, together with LBP and hog we can get state-of-the-art recognition. Together with the segmentation algorithms, we can do semantic segmentation. So if we have a key-point descriptor like BRIEF, this enables us to cover quite a lot of the popular computer vision applications. Just saying ;) LBP can also be used but I think BRIEF works better for natural images... Cheers, Andy ----- Ursprüngliche Mail ----- Von: "Stéfan van der Walt" <stefan(a)sun.ac.za> An: scikits-image(a)googlegroups.com Gesendet: Donnerstag, 9. August 2012 06:31:58 Betreff: EuroSciPy sprint Hi, everyone We'll be sprinting at EuroSciPy, and would like to solicit ideas. Currently, the sprint description reads as follows. I'd love to hear other suggestions! """ scikits-image is a collection of algorithms for image processing. See http://skimage.org for more detail. Ideas for the sprint: - Add text, anti-aliasing to the draw module - Integrate Tony's visualization tools - Add binary features (BRIEF, BRISK, FREAK) - Update the user guide - Better video loading - Hough transform parameter space explorer Outstanding challenges: - Blurring kernel estimation: http://bit.ly/Nril3u - http://stackoverflow.com/questions/10196198/how-to-remove-convexity-defects… And write up this one as an example: - http://stackoverflow.com/questions/8686926/python-image-processing-help-nee… """ Regards Stéfan

4 4

Re: Alpha-expansion, alpha-beta swaps and patents
by Andreas Mueller 15 Aug '12

15 Aug '12

Hi Stefan. I fully agree with you. This is why I advertised BRIEF earlier. The thing is: it might prove very hard to find a replacement for alpha expansion. I'll be in Vienna next week and I'll talk to Vladimir Kolmogorov and see if I can get his TRW-S code, but that probably belongs to microsoft. And I'm not sure it is humanly possible to reimplement Vladimirs algorithms ;) Just to recall why I want a graph cut method: That basically is state of the art in small baseline stereo, optical flow, multi-class image segmentation, interactive segmentation, ... Cheers, Andy On 08/15/2012 03:17 AM, StÃ¯Â¿Â½fan van der Walt wrote: > Hi Andy > > On Tue, Aug 14, 2012 at 12:31 PM, Andreas MÃ¯Â¿Â½ller > <amueller(a)ais.uni-bonn.de> wrote: >> To restart a conversation we had before on SIFT: >> Do we want patented algorithms in sklearn? > While there's nothing stopping us from implementing patented > algorithms, I do not like the implications. It means we have to start > advertising to users that there are specific functions that they > cannot use without carefully examining the conditions. I'd much > prefer for the whole package to be freely usable. > > I get very unhappy when I see patents on algorithms such as SIFT, > built on top of (or at least inspired by) an entire volume of > non-patented research. Fortunately, there are often > work-arounds--e.g., I believe the STAR (CenSurE) features used in > OpenCV (BSD-licensed!) are not patented, and work as well as or better > than SIFT. So I suggest we implement that instead: > > http://pr.willowgarage.com/wiki/Star_Detector > > StÃ¯Â¿Â½fan

1 0

Alpha-expansion, alpha-beta swaps and patents
by Andreas Müller 15 Aug '12

15 Aug '12

Hi everybody. To restart a conversation we had before on SIFT: Do we want patented algorithms in sklearn? I just found out that alpha-expansion and alpha-beta swaps are patented, which makes me very unhappy. What do patents mean: We can implement patented algorithms and distribute them under BSD license. If someone uses the software for commercial purposes, they can be sued by the patent holder. If we choose to include patented algorithms, we can include them with a warning. Then there is no problem in having SIFT or including graph cuts with alpha expansion. Cheers, Andy

2 1

Re: EuroSciPy sprint
by Johannes Schönberger 14 Aug '12

14 Aug '12

Could you paste the code somewhere or send it to me via EMail? Would be great.

2 1

Re: EuroSciPy sprint
by Johannes Schönberger 14 Aug '12

14 Aug '12

If you haven't already done so, also have a look at the linked PDF document on this site.

1 0

Re: EuroSciPy sprint
by Johannes Schönberger 14 Aug '12

14 Aug '12

Hi, would love to join you but unfortunately I am not at EuroScipy. The anti-aliasing was on my agenda for the next weeks, but if you want to do it now - here is a good paper I found recently: http://free.pages.at/easyfilter/bresenham.html Am Donnerstag, 9. August 2012 07:31:58 UTC+2 schrieb Stefan van der Walt: > > Hi, everyone > > We'll be sprinting at EuroSciPy, and would like to solicit ideas. > Currently, the sprint description reads as follows. I'd love to hear > other suggestions! > > """ > > scikits-image is a collection of algorithms for image processing. See > http://skimage.org for more detail. > > Ideas for the sprint: > > - Add text, anti-aliasing to the draw module > - Integrate Tony's visualization tools > - Add binary features (BRIEF, BRISK, FREAK) > - Update the user guide > - Better video loading > - Hough transform parameter space explorer > > Outstanding challenges: > > - Blurring kernel estimation: http://bit.ly/Nril3u > - > http://stackoverflow.com/questions/10196198/how-to-remove-convexity-defects… > > And write up this one as an example: > > - > http://stackoverflow.com/questions/8686926/python-image-processing-help-nee… > """ > > Regards > Stéfan >

2 2

Re: EuroSciPy sprint
by jeff witz 14 Aug '12

14 Aug '12

Hello, May one or two examples of Image Registration could be a good idea ? Le jeudi 9 août 2012 07:31:58 UTC+2, Stefan van der Walt a écrit : > > Hi, everyone > > We'll be sprinting at EuroSciPy, and would like to solicit ideas. > Currently, the sprint description reads as follows. I'd love to hear > other suggestions! > > """ > > scikits-image is a collection of algorithms for image processing. See > http://skimage.org for more detail. > > Ideas for the sprint: > > - Add text, anti-aliasing to the draw module > - Integrate Tony's visualization tools > - Add binary features (BRIEF, BRISK, FREAK) > - Update the user guide > - Better video loading > - Hough transform parameter space explorer > > Outstanding challenges: > > - Blurring kernel estimation: http://bit.ly/Nril3u > - > http://stackoverflow.com/questions/10196198/how-to-remove-convexity-defects… > > And write up this one as an example: > > - > http://stackoverflow.com/questions/8686926/python-image-processing-help-nee… > """ > > Regards > Stéfan >

2 1