Remarks for transform.warp function

[Nadav Horesh]

2012/5/16 Tony Yu <tsyu80 at gmail.com>

>
>
> On Wed, May 16, 2012 at 1:45 AM, Nadav Horesh <nadavh.horesh at gmail.com>wrote:
>
>> It solves the problem.
>> What do you think about letting enabling output_shape=(m,n) even for a
>> colour image, implicitly assuming that the 3rd dimension is copied from the
>> original image.
>>
>>   Nadav
>>
>
> Currently, color images work fine when no `output_shape` is specified. I'm
> not sure what you mean by "implicitly assuming that the 3rd dimension is
> copied from the original image". Are you suggesting that for color images
> setting `output_shape=(m, n)` should actually return a shape of `(m, n,
> 3)`? If that's the case, I strongly disagree because it seems
> magical---i.e. it's contradicting the user's command.
>
> Actually, I don't understand the use case for `output_shape`, so I could
> be missing something...
>
> -Tony
>
> Nadav: could you bottom-post or interleave posts<http://en.wikipedia.org/wiki/Posting_style> to
> the list---it helps when following long conversations.
>
>
>>
>> 2012/5/15 Tony Yu <tsyu80 at gmail.com>
>>
>>>
>>>
>>> On Tue, May 15, 2012 at 2:28 PM, Nadav Horesh <nadavh.horesh at gmail.com>wrote:
>>>
>>>> def shift(xy):
>>>>     return xy+10
>>>>
>>>> from skimage.transform import warp
>>>> from skimage.transform import warp
>>>>
>>>> camera = samples.camera()
>>>> warp(camera, shift, output_shape=camera.shape)
>>>>
>>>>
>>>> ---------------------------------------------------------------------------
>>>> ValueError                                Traceback (most recent call
>>>> last)/home/nadav/pyscripts/<ipython-input-11-ce1e92562ac8> in <module>()
>>>> ----> 1 warp(camera, shift, output_shape=camera.shape)
>>>> /usr/lib64/python3.2/site-packages/scikits_image-0.6dev-py3.2-linux-x86_64.egg/skimage/transform/_warp.py
>>>> in warp(image, reverse_map, map_args, output_shape, order, mode, cval)
>>>>      92
>>>>      93     # Place the y-coordinate mapping
>>>>
>>>> ---> 94     _stackcopy(coords[1, ...], tf_coords[0, ...])
>>>>      95
>>>>      96     # Place the x-coordinate mapping
>>>>
>>>> /usr/lib64/python3.2/site-packages/scikits_image-0.6dev-py3.2-linux-x86_64.egg/skimage/transform/_warp.py
>>>> in _stackcopy(a, b)
>>>>      23
>>>>      24     """
>>>> ---> 25     a[:] = b[:, :, np.newaxis]
>>>>      26
>>>>      27 def warp(image, reverse_map, map_args={},
>>>> ValueError: output operand requires a reduction, but reduction is not
>>>> enabled
>>>>
>>>>
>>>>   Nadav
>>>>
>>>
>>>
>>> Hi Nadav,
>>>
>>> Could you check to see if the fix in my fix-stack-copy solves your
>>> issues (it should fix the example you gave above, but it sounds like you
>>> might have other  uses). You can either make changes<https://github.com/tonysyu/scikits-image/commit/048c6c06ebe9935aa6844a09ed85396c156befbd>to your local copy, or you can create a new branch based off of mine. I
>>> think you mentioned that you're new to git so...
>>>
>>> Add my repo as a remote:
>>>     $ git remote add tonysyu git at github.com:tonysyu/scikits-image.git
>>> Fetch my branches:
>>>     $ git fetch tonysyu
>>> Create and checkout a new branch based off of my branch:
>>>     $ git checkout -b fix-stack-copy tonysyu/fix-stack-copy
>>>
>>> (Let me know if you have issues with the above---I'm working from
>>> memory, so it may be flawed)
>>>
>>> Best,
>>> -Tony
>>>
>>>
>>>>
>>>>
>>>>
>>>>
>>>> 2012/5/15 Stéfan van der Walt <stefan at sun.ac.za>
>>>>
>>>>> Hi Nadav
>>>>>
>>>>> On Tue, May 15, 2012 at 7:24 AM, Nadav Horesh <nadavh.horesh at gmail.com>
>>>>> wrote:
>>>>> > Remarks for today's skimage snaphot:
>>>>> >
>>>>> > The documentation state that the input image is a 2D array. In
>>>>> practice it
>>>>> > can be a colour image.
>>>>> > I got failures whenever I used the output_shape parameter
>>>>>
>>>>> What code are you referring to?
>>>>>
>>>>> Stéfan
>>>>>
>>>>
>>>
>
The warp mapping usually define an image of a different size than the
original. The common case is that the distortion corrected image is cliped
to the original image size, this is why that this is the default. However,
one may want to see the full details of the original image, and thus
specify output_shape > image.shape.

You got me right about the implicit dimension. The logic: The image size is
the number of pixels, disregarding the colour depth. You are right if you
consider the image as an array of values, but if you consider it as an
image, the image shape is just the first 2 dimensions.  Anyway it is not
that important.

  Thank you,

    Nadav.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-image/attachments/20120516/d29c4758/attachment.html>