I've found that within the universe of SciPy +extensions (like the various SciKits, really anything that likes NumPy arrays) it's fairly easy to get things to play nice without much trouble. Interlacing algorithms from there with OpenCV, SimpleITK, VTK, and the like are more of a chore. They all exist in their own worlds as far as internal datatypes, syntax structure, etc. Nowhere near as friendly to hack together (it can be done, with order(s) of magnitude more effort). However, certain packages in the SciPy universe use these, usually wrapped such that the user never sees it. E.g. Mayavi uses VTK heavily, and Scikit-Image now has a SimpleITK plugin for io - but they're wrapped to users only have to deal with Python/NumPy. In my opinion there is definite value in ports, though they should probably take somewhat lower priority than new/novel work. On Thursday, April 18, 2013 2:24:29 PM UTC-5, Ronnie Ghose wrote:
So first off is there a sort of to-do list? Secondly, should scikit-image be used in conjunction with OpenCV? I thought in general it would be standalone like a few of the other scikits or numpy/scipy (standalone in the sense of only using other python libraries as dependencies).
On Thu, Apr 18, 2013 at 12:50 PM, Zachary Pincus <zachary...@yale.edu<javascript:>
wrote:
Would this be a useful thing to add then? I thought it was interesting you made references to OpenCV rather then internal options.
Yeah, but no need to reinvent the wheel. OpenCV is really focused on algorithms for cameras-moving-around-in-the-world, which big chunks of your problem are. So if you want to port some of that code over to skimage, that would be a cool and useful contribution! But if you just wanted to solve your problem ASAP, then OpenCV might be your best resource...
Zach
PS. Fun story: a while ago I got a sparse optical-flow algorithm (using cross-correlation) running on a GPU. It was totally non-portable (pre-CUDA, custom-shader code), and it didn't actually work precisely right, but I think that it's something that shouldn't be too much trouble to actually make happen these days...
On Thu, Apr 18, 2013 at 8:19 AM, Stéfan van der Walt <ste...@sun.ac.za<javascript:>> wrote: On Wed, Apr 17, 2013 at 11:16 PM, Zachary Pincus <zachary...@yale.edu <javascript:>> wrote:
Identifying features to track between frames is a pretty classic task in computer vision. Things that come to mind would include the simple Harris corner detector (in skimage), or more recent things like SIFT features and its variants like SURF etc. (DAISY is in skimage) might be of use. All of these tools should be in OpenCV too, IIRC. (Time to spend a while on wikipedia reading about these things and related, if it's all totally unfamiliar.)
Having binary features (there are many) and Star Features (also known as "Consensus") in skimage would be very valuable.
Alternately, you could try optical flow based techniques to densely track local pixel patches between frames.
That's also a fun application, and a good way to learn Cython!
Stéfan
-- You received this message because you are subscribed to the Google Groups "scikit-image" group. To unsubscribe from this group and stop receiving emails from it, send an email to scikit-image...@googlegroups.com <javascript:>. For more options, visit https://groups.google.com/groups/opt_out.
-- You received this message because you are subscribed to the Google Groups "scikit-image" group. To unsubscribe from this group and stop receiving emails from it, send an email to scikit-image...@googlegroups.com <javascript:>. For more options, visit https://groups.google.com/groups/opt_out.
-- You received this message because you are subscribed to the Google Groups "scikit-image" group. To unsubscribe from this group and stop receiving emails from it, send an email to scikit-image...@googlegroups.com <javascript:>. For more options, visit https://groups.google.com/groups/opt_out.