help-gift
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [help-GIFT] Optimizing


From: risc
Subject: Re: [help-GIFT] Optimizing
Date: Fri, 18 Aug 2006 11:35:44 -0500
User-agent: Mutt/1.4.1i

good enough. ;)

As i stated earlier, in hindsight, i agree with the issue David pointed out, 
and I'll submit a patch for that shortly. none of my work should require hard 
codings. that was just laziness on my part, trying to make the shortest 
possible patches. so theres not so much to read. :)

Is for the Perl hacks, I'm willing to try my hand, tho i am not a Perl, or a 
c++ programmer. ;)

If we end up reverting my patch sets, thats fine, as the process we're going 
through puts me in an ideal position to maintain a "feature-extractor-fast" 
version. you were both right in insisting i break my patch up into small edible 
pieces, even if they don't go in the "academic-simple" version of the tree. my 
hacks are non-obvious, and bound to confuse a researcher who wants to change 
something, without having a full understanding of the intricacies of the C 
programming language. Hell, they confuse me sometimes. :)

so, my weekend project list is as follows:
continue pulling patches from my 'private' tree,
attempt to add code to the Perl script to select a feature extractor,
attempt to pull patches from my version of gift-add-collection. specifically, 
allow for "pre-seeding" of the feature extraction, and less calls to one of the 
finishing programs...

I'll be applying the 70-* and 80-* patches sometime today or tomorow, unless 
anyone screams.

I've done submitted the test procedure i use for testing the feature extractor. 
any comments?

Julia Longtin <address@hidden>


On Fri, Aug 18, 2006 at 06:10:55PM +0200, Wolfgang Mueller wrote:
> Dear Julia, Dear David, diear all,
> 
> I am not able to seriously review patches in the next few days (I did
> say so, right ;-) ). However, though I am not a pure fan of the kind of deal
> proposed below, I suggest the following.
> 
> 1) we accept your patches, with the possibility of later changing the
> most hairraising hardcodings
> 2) someone do some hacks to my perl (possibly with my help) to make
> multiple feature extactors peacefully coexist.
> 3) if it turns out that your patches are really bad for some purposes
> and cannot be cured without losing the speed advantages, we add a more
> conservative version back to the tree.
> 
> Sounds good to both/all of you?
> 
> Cheers,
> Wolfgang
> 
> On Fri, 18 Aug 2006, address@hidden wrote:
> 
> > I couldn't agree more.
> > 
> > Coding is my way of stress-releif, from my normal day-to-day activities.
> > 
> > My next sets of optimizations are more algorithmic changes inside of 
> > gabor_filter, and therefore will benifit both the multi-extractor, and the 
> > single-extractor. I hoped to get these committed soon, but i'm not yet sure 
> > if i should even apply my 70* and 80* patches, due to the aparent 
> > controversy of my earlier patches. 
> > 
> > can i get the live list members to look over my patches, and see what you 
> > think? 
> > 
> > I'm perfectly willing to start work on the multi-extractor, as soon as my 
> > present work is in. this represents MONTHS of my effort, and i'd really 
> > hate to see it discarded due to 'accademic cleanliness' concerns. coding 
> > cleanliness? i'll agree with that all day long. ;)
> > 
> > I've got a big weekend of gift hacking ahead of me. please pick a 
> > direction, and i'll follow it. ;)
> > 
> > Julia Longtin <address@hidden>
> > 
> > 
> > On Fri, Aug 18, 2006 at 09:32:34AM +0200, address@hidden wrote:
> > > Well, the feature extraction as is has suffered from the fact that it is 
> > > a quick perl hack from the late nineties made to simplify the usage of a 
> > > command line tool for single images. My tries to improve this in a more 
> > > fundamental manner got stuck when I changed my job and realised that 
> > > coding now takes too much of my time. I did not like the alternatives.
> > > 
> > > My view now is that the project could greatly profit if we got something 
> > > going that is good enough. Yes, I do think making GIFT work for any-sized 
> > > images is something useful, and it should not be stopped. I do not think 
> > > my proposition is in any way in conflict with that.
> > > 
> > > Doing feature extraction with multiple files via command line parameters 
> > > appears useful, however you won't get your 90k images in the command 
> > > line. Hmm. There is xargs.
> > > 
> > > Anyway my feeling is that the pipe thingy will be more flexible and 
> > > useful. It is a gut feeling.
> > > 
> > > Cheers,
> > > Wolfgang   
> > > 
> > > 
> > > 
> > >  
> > >                    
> > > 
> > > 
> > > _______________________________________________
> > > help-GIFT mailing list
> > > address@hidden
> > > http://lists.gnu.org/mailman/listinfo/help-gift
> > 
> > 
> > _______________________________________________
> > help-GIFT mailing list
> > address@hidden
> > http://lists.gnu.org/mailman/listinfo/help-gift
> --
> Dr. Wolfgang Mueller
> LS Medieninformatik
> Universitaet Bamberg




reply via email to

[Prev in Thread] Current Thread [Next in Thread]