I noticed a documentation inconsistency in `labels` for random_walker; it hadn't been updated with the rest of the file and still said `labels` should be the same shape as `data`. This is true for the single-channel case, but for multichannel input `labels` should look like a single channel (missing the final dimension in `data`). Proposed docfix is submitted in a PR.
Wow, that was quick :-), thanks!
I'll have a look tomorrow, I'm just back from Brussels and Euroscipy!
On Mon, Aug 27, 2012 at 02:18:22PM -0700, Josh Warner wrote:
> Pull request is up! ï¿½I also fixed a bug in the fallback call of
> _solve_cg() if pyamg is not installed. ï¿½Comments appreciated.
> On Monday, August 27, 2012 7:55:11 AM UTC-5, Emmanuelle Gouillart wrote:
> Hi Josh,
> my changes have been merged, so if you pull from github they are in
> the master branch now, you can add your changes and make a pull request!
> On Thu, Aug 23, 2012 at 02:50:05PM -0700, Josh Warner wrote:
> > ï¿½ ï¿½I'd be happy to include your suggested changes, if it's simpler to
> go that
> > ï¿½ ï¿½route than have two separate pull requests. ï¿½I don't want to step
> on any
> > ï¿½ ï¿½toes regarding attribution, though. ï¿½
> > ï¿½ ï¿½Having the 'soft' probabilities available is definitely a useful
> > ï¿½ ï¿½improvement (cleanup, fuzzy post-processing), and I'm glad you
> > ï¿½ ï¿½that! ï¿½
> > ï¿½ ï¿½Josh
> > ï¿½ ï¿½On Monday, August 20, 2012 4:11:45 PM UTC-5, Emmanuelle Gouillart
> > ï¿½ ï¿½ ï¿½Hi Josh,
> > ï¿½ ï¿½ ï¿½extending the random walker algorithm to multichannel images
> sounds like
> > ï¿½ ï¿½ ï¿½a very nice idea, and I'm sure that it would be useful for other
> > ï¿½ ï¿½ ï¿½(including myself!).
> > ï¿½ ï¿½ ï¿½In fact, I had started working on some improvements of the random
> > ï¿½ ï¿½ ï¿½code but didn't go as far as proposing the pull request. I have a
> > ï¿½ ï¿½ ï¿½for that on
> > ï¿½ ï¿½
> > ï¿½ ï¿½ ï¿½maybe
> > ï¿½ ï¿½ ï¿½you can include these changes when you contribute your changes?
> > ï¿½ ï¿½ ï¿½Cheers,
> > ï¿½ ï¿½ ï¿½Emmanuelle
> > ï¿½ ï¿½ ï¿½On Mon, Aug 20, 2012 at 10:03:53AM -0700, Josh Warner wrote:
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½I have modified the existing random walker algorithm
> into a fully
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½backwards-compatible version which allows inclusion of
> > ï¿½ ï¿½ ï¿½multispectral data,
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½e.g. RGBA channels or different (registered) image
> modalities. ï¿½I
> > ï¿½ ï¿½ ï¿½really
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½liked the existing algorithm, so I just extended it
> rather than
> > ï¿½ ï¿½ ï¿½write one
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½from scratch for my own purposes. ï¿½The overhead is
> > ï¿½ ï¿½ ï¿½multispectral
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½processing is triggered if data is passed as an iterable
> of arrays
> > ï¿½ ï¿½ ï¿½rather
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½than just an array. ï¿½
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½This amounts to combining image gradients as
> > ï¿½ ï¿½ ï¿½and
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½dividing by sqrt(#channels). ï¿½For obvious reasons, the
> > ï¿½ ï¿½ ï¿½channels
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½must be pre-processed to have data on similar ranges by
> > ï¿½ ï¿½ ï¿½or a
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½similar method. ï¿½Not usually a problem for RGB, but in
> > ï¿½ ï¿½ ï¿½imaging
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½this rears its head.
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½Would this be of interest to the community? ï¿½I'd be
> happy to
> > ï¿½ ï¿½ ï¿½contribute
> > ï¿½ ï¿½ ï¿½> ï¿½ ï¿½the changes if there is interest.
> > References
> > ï¿½ ï¿½Visible links
> > ï¿½ ï¿½1.
> Visible links
> 1. https://github.com/emmanuelle/
> 2. https://github.com/emmanuelle/