[3dem] [ccpem] lost gain reference image
Carlos Oscar S. Sorzano
coss at cnb.csic.es
Tue Dec 4 13:56:36 PST 2018
Dear Marin,
thank you for the reminder, since I had certainly forgotten. What you
call a contrast reversal is most likely a nomenclature issue. If you
have two images, I_observed and I_ideal, and the relationship between
the two is simply a multiplicative term
I_observed = H I_ideal
exactly the same relationship is described by the relationship (assuming
there are no zeros)
F I_observed = I_ideal
The relationship between F and H is F=1/H, which implies a contrast
reversion. I have explictly avoided G to denote any of these two images
(F or H) to avoid confusions. Which of the two is called "gain" is a
matter of definition. I prefer describing H as the gain, but I
understand other people may prefer the other option.
Regarding R^2, it is true I did not define it assuming it was general
knowledge. It is a very standard statistical measure of quality called
the coefficient of determination
(https://en.wikipedia.org/wiki/Coefficient_of_determination), and it
expresses the fraction of the original variance explained by a model. If
R^2=1, the model has totally explained the original variance, while if
R^2=0, the model explains the same variance as the mean of the
observations, which is the simplest, sensible model we could have.
Cheers, Carlos Oscar
El 04/12/2018 a las 21:48, Marin van Heel escribió:
> Hi Carlos Oscar!
> I just remembered I had posed a question about your camera
> normalisation paper ( https://www.ncbi.nlm.nih.gov/pubmed/29551714.)
> on this site some two months ago, in which you critisized our 2015
> camera normalisation paper
> (https://www.nature.com/articles/srep10317). Did you already respond
> to my question and I missed your answer?
>
> Cheers
> Marin
>
> My question was:
>
> QUOTE:
> What you call our "gain image" is - apart from an erroneous contrast
> reversal - actually more similar to the "official" gain image in your
> Fig 1 than does the one generated with your proposed algorithm. I
> would be interested in knowing what the R2 turns out to be after you
> correct the contrast reversal since it visually is better than yours.
> It would be nice if you could respond to this mailing including that
> information!? By the way how exactly is this R2 metric defined (I
> could not find it anywhere in the paper)?
> END QUOTE
>
> On Tue, Oct 2, 2018 at 10:21 PM Marin van Heel
> <marin.vanheel at googlemail.com <mailto:marin.vanheel at googlemail.com>>
> wrote:
>
> Dear Carlos Oscar and Dimitry,
>
> Unfortunately, you seem to have missed the point of our Afanasyev
> 2015 paper. Our paper does not try to duplicate the
> "experimentally determined Gain image" but tries to normalize the
> signal from each pixel to the same average and the same standard
> deviation at the exposure and contrast level that the data set was
> recorded. Our approach typically improves significantly on
> standard "/a priori/" flat field/gain corrections.
>
> We are not directly interested in generic "gain images" as such
> and we certainly do not generate "gain images" that have an
> inverted contrast when compared to the other ones you have in
> Figure #1 of your paper. Your comments on our methods are thus not
> appropriate: "/T//o the best of our knowledge, the only article
> that addresses a similar problem is that of Afanasyev //et
> al.//(2015). In their work, they assimilate the gain of the camera
> to the standard deviation of each pixel over a large number of
> movies, and they prove this is a successful way of identifying
> dead pixels. However, our results show that this approach does not
> provide a consistent gain estimation (Fig. 1)/."
>
> What you call our "gain image" is - apart from an erroneous
> contrast reversal - actually more similar to the "official" gain
> image in your Fig 1 than does the one generated with your proposed
> algorithm. I would be interested in knowing what the R2 turns out
> to be after you correct the contrast reversal since it visually is
> better than yours. It would be nice if you could respond to this
> mailing including that information!? By the way how exactly is
> this R2 metric defined (I could not find it anywhere in the paper)?
>
> I would want to suggest you and your colleagues to use the FRC
> metric to prove that your approach does indeed remove the
> influence of the various patterns of your detectors exhibits.
>
> My two cents
>
> Marin
>
> =====================
>
>
> On 02/10/2018 15:19, Carlos Oscar Sorzano wrote:
>> By the way, in our article we compared both methods (ours and
>> Marin).
>>
>> Kind regards, Carlos Oscar
>>
>>
>> On 01/10/2018 21:23, Marin van Heel wrote:
>>> Dear Da,
>>>
>>> In IMAGIC-4D you can perform the necessary camera correction!
>>> (https://www.nature.com/articles/srep10317). It does it better
>>> than any manufactures correction and improves the data
>>> significantly even when performed after using the standard gain
>>> correction.
>>>
>>> Cheers,
>>>
>>> Marin
>>>
>>>
>>> =====================================================
>>>
>>> On 01/10/2018 15:36, Da Cui wrote:
>>>> Hi all,
>>>> The gain reference image for one dataset was missing by
>>>> accident. In order to achieve a more accurate motioncor result,
>>>> does anyone have idea about how to generate a gain reference
>>>> image from the dataset (around 3k movies)?
>>>> Thank you so much for your help!!!
>>>> ---Da
>>>>
>>>> ########################################################################
>>>>
>>>>
>>>> To unsubscribe from the CCPEM list, click the following link:
>>>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
>>>
>>>
>>
>
> --
> ==============================================================
>
> Prof Dr Ir Marin van Heel
>
> Laboratório Nacional de Nanotecnologia - LNNano
> CNPEM/LNNano, Campinas, Brazil
>
> tel: +55-19-3518-2316
> mobile +55-19-983455450 (current)
> mobile +55-19-981809332
> (041-19-981809332 TIM)
> Skype: Marin.van.Heel
> email: marin.vanheel(A_T)gmail.com <http://gmail.com>
> marin.vanheel(A_T)lnnano.cnpem.br <http://lnnano.cnpem.br>
> and: mvh.office(A_T)gmail.com <http://gmail.com>
>
> --------------------------------------------------
> Emeritus Professor of Cryo-EM Data Processing
> Leiden University
> Mobile NL: +31(0)652736618 (ALWAYS ACTIVE SMS)
> --------------------------------------------------
> Emeritus Professor of Structural Biology
> Imperial College London
> Faculty of Natural Sciences
> email: m.vanheel(A_T)imperial.ac.uk <http://imperial.ac.uk>
> --------------------------------------------------
>
> I receive many emails per day and, although I try,
> there is no guarantee that I will actually read each incoming email.
>
--
------------------------------------------------------------------------
Carlos Oscar Sánchez Sorzano e-mail: coss at cnb.csic.es
Biocomputing unit http://i2pc.es/coss
National Center of Biotechnology (CSIC)
c/Darwin, 3
Campus Universidad Autónoma (Cantoblanco) Tlf: 34-91-585 4510
28049 MADRID (SPAIN) Fax: 34-91-585 4506
------------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20181204/4782eae1/attachment-0001.html>
More information about the 3dem
mailing list