|
Here is a new version of the script published with the authorization of Martin Krzywinski: basically, it acts as glue code between Flickr and his web service to generate four color-related tags -- see the script to know more.
Posted 77 months ago.
(permalink)
|
|
I suggest you post an "image" in your syfou3 photostream that explains (either in the image, in the description, or both) what this is all about. I'm sure many people will head there to find out why they've received these cryptic tags. You might want to put the same explanation in the syfou3 profile as well.
Posted 77 months ago.
(permalink)
|
|
|
That's an excellent suggestion, Orrin! I will do it later today.
Posted 77 months ago.
(permalink)
|
|
this is very interesting syfou3 - all i ask is that you take it slow and make sure you get feedback often.
i just want to make sure people are not weirded out, pretty much what orrin said.
i like it though!
Posted 77 months ago.
(permalink)
|
|
Ah, that's what these strange tags appearing on my pictures are !!!
Posted 77 months ago.
(permalink)
|
|
I've read through the posts on this new tagging, getting the answer to why the strange tags appeared on my photos, but not one has explained why this is a good idea. Tags are so people can find photos. Are people really searching for photos by specific RGB or hex values? And if so, WHY?
I AM weirded out by this, hexod. What's the point? More techno-jargon or a real useful tool for some purpose I'm not understanding? Please enlighten me.
Posted 77 months ago.
(permalink)
|
|
wizmo: the point is that those tags make it much easier for coders to build experiments or applications that need the average color values of images.
I'm sure you know krazydad.com/colrpickr/ - in order to build this all shown images had to be downloaded first, their average colour had to be calculated and stored in a separate database. This information is now already contained in the tags, which saves bandwidth and time.
Searching for specific RGB values. Hmmm. Probably not the first thing someone would do, but why not? Like "find an image that fits your site's background color?"
Posted 77 months ago.
(permalink)
|
|
|
I guess on of the important use of Color Fields are to build well harmonized backgrounds, photo mosaic and other similar things: those tags simply relieve everyone from having to compute the average information for themselves when they need it.
All tags on Flickr are not directly searchable; see for instance those related to Geo Tagging...
Posted 77 months ago.
(permalink)
|
|
I think this is great work and really adds value to photos shared in the pool. Once you're over the hill of tagging the archive, you could probably hang off the RSS feed to only update photos as they're added to the pool.
Posted 77 months ago.
(permalink)
|
|
Lovely idea! Alot of applications will be able to use such tags in so many ways!
Posted 77 months ago.
(permalink)
|
|
|
it is a pretty cool idea, although i just noticed a concern...
syfou3: i think it would also be a good idea to take it slow because otherwise the script my start to clutter people's recent comments.
just a concern. otherwise, this is pretty cool man.
Posted 77 months ago.
(permalink)
|
|
i reckon once one or more cool applications are using this (come on jbum), then it'll make more sense...
Posted 77 months ago.
(permalink)
|
|
|
Thanks, I didn't know the crazydad site, and it is cool.
I'm still not too happy about all these tags full of techno-gibberish cluttering up my page and comment box ( yes, hexod is right, the tags have taken up tons of my comment space) and I really wouldn't like it if people used my photos for commercial purposes without permission, so making it easier doesn't have a lot of appeal.
I'm going to remove the tags. It this means I get kicked out of the group, so be it, do what you must, but thanks for the explanations, and best wishes.
Posted 77 months ago.
(permalink)
|
|
I really appreciate it when smart people get more effort out of pictures, flickr and all the rest. thumbs up!
Posted 77 months ago.
(permalink)
|
|
|
Maybe there has to be an additional tag people can add to their images if they don't want them automatically tagged?
Like "nobottagshere"?
Posted 77 months ago.
(permalink)
|
|
I already thought there was a way for people to turn off other people being able to tag their photos ... so are they saying they don't mind other people adding tags, but just not these?
Posted 77 months ago.
(permalink)
|
|
putting a photo in a group overrides any privacy settings on that photo, including tagging. If it's in a group, any group member can tag it. That's how groups work.
Posted 77 months ago.
(permalink)
|
|
|
wizmo: If you really do not want the tags, all you have to do is blocking me -- tags should even be automatically removed; if you don't, your photos in colorfields will just get tagged over and over anyway.
Of course, every time someone does so, it removes a little appeal to the whole effort, but I do respect your point of view (although not sharing it) and I do not want to make your flickr experience less enjoyable over some simple tags.
Originally posted 77 months ago.
(permalink)
syfou3 edited this topic 77 months ago.
|
|
|
Well in my book tagging is goodness, and the more relevant tags the better. wizmo, I'm saddened you want to remove the tags, but i respect your opinion. I'm guessing syfou3 created a dedicated account just for this instance so you can block him.
For my part, I'd welcome a more coarse-grained tags as well as the ones being added tonight - "red", "green", "brown" etc, that way we'd see clusters - i notice very few of mine have other photos with the same RRGGBB values..
Posted 77 months ago.
(permalink)
|
|
|
good point psd. course grained would be really really really awesome. not just the clusters that would develop, but the fact that it makes more human readable metadata.
Originally posted 77 months ago.
(permalink)
hexodus... (a group admin) edited this topic 77 months ago.
|
|
|
Just so you all know, I have just finished tagging the whole group (it took around height hours). The latest version of the script used can be found here.
I intend tagging new photos once a day (starting at 23h00 GMT), and to re-check the whole group at the same time twice a month -- I cannot do more with my current bandwidth.
Originally posted 77 months ago.
(permalink)
syfou3 edited this topic 77 months ago.
|
|
I have only one photo in this pool and I just saw the tags on it and came here on a hunch. Metadata is cool!
Posted 77 months ago.
(permalink)
|
|
I'm happy to be tagged for this (and looking forward to seeing what use the tags are put to. Keep us up to date!).
Posted 77 months ago.
(permalink)
|
|
Yay for metadata!
Posted 77 months ago.
(permalink)
|
|
PSD -- "very few" of your photos have matches with exact RGB values? I'd imagine that the chances of any photo being an exact match of another are very slim indeed.
Either way -- finding approximations would be cool. I don't think more "coarse-grained" tags would be the best answer: more intelligent search would be far better. Clustering apart, tag searches basically match alphanumeric strings. A more intelligent search tool could be looking for values within specified ranges.
Posted 77 months ago.
(permalink)
|
|
|
A few months back, I wrote such a search... You could be surprised about the images distribution -- look at the graph: it is not that uncommon to have images at less than 1/256th units of euclidean distance in RGB spaces: it was the case for about 300 images belonging to the group last December.
Originally posted 77 months ago.
(permalink)
syfou3 edited this topic 77 months ago.
|
|
|
ark: "any photo being an exact match of another are very slim indeed.", er, something in the order of 1 in 16777216, assuming even distribution, I'd guess :-)
ark "A more intelligent search tool could be looking for values within specified ranges."
I think a regex search for tags would be cool, and useful for geotagging, but can imagine the difficulties how to implement such searches efficiently.
Posted 77 months ago.
(permalink)
|
|
|
Have a look here:
This is the distribution of the nearest neighbor distance in RGB spaces for the Color Fields group, as measured by me last December. As you see, images are anything but evenly distributed. ;-)
In fact, if you look at it close enough, it means that on average, Euclidean distance between two images in the pool is around 14 1/256th of units. This has one important repercussion on search:
it means that no matter how well you optimize things, you would have on average around 4/3*pi*14^3 (approx 11500) queries to make on Flickr for finding the nearest image from another...Do I need to mention there are less than 5000 images in total in the group?
So yes, we need something coarser if we want direct search...
Posted 77 months ago.
(permalink)
|
|
|
As for something coarser - it shouldn't be difficult to write a script that uses the information from the average or median color tag and then looks for the closest matching html color name (http://www.w3schools.com/HTML/html_colornames.asp) and then - I almost do not dare to mention it - automatically adds that as a another tag.
Posted 77 months ago.
(permalink)
|
|
wow, I just checked back to flickr and saw all the tags on my images, 110 pages of them. You did it syfou3, good or not. Maybe colorfields will benefit from not only the tagging but the traffic of confusion.
Posted 77 months ago.
(permalink)
|
|
|
Well, cudmore, you are the second biggest image contributor to this group (110 images)-- you were bound to be one of the most impacted...I hope it will not cause to much undesirable traffic: after all, it is merely 84 bytes of new information per image, hence only 9 KB for all of yours...
Posted 77 months ago.
(permalink)
|
|
|
Quasimodo: I would be afraid that using web colors would not be good enough (there is only approx. 140 of them)... I did a little experimentation (351 KB) using my X.org color definitions; let me know what you think.
Posted 77 months ago.
(permalink)
|
|
|
syfou3: hey, that was quick. But you are right - some of them are way off. Which values did you use to determine the shortest distance? The average_rgb or average_hsv? I assume you calculated the distance in HSV space, but here's is why I'm asking:
From the method the various average/median values are calculated I think that converting the average_hsv value to RGB will not result in the same value as average_rgb contains, is that possible? The average_hsv value makes only sense if I'm interested in the average hue alone or the average saturation or brightness.
So in order to determine the correct distances I'd first convert the average_rgb value to hsv as well as the html-colors and then try to find the nearest neighbours.
Hmm, does my babbling make any sense to you?
Posted 77 months ago.
(permalink)
|
|
|
For this, I minimised an Euclidean metric in RGB space, using the computed average.
Is the metric minimisation a space invariant? Yes, as long as the transform between spaces is reversible (by definition, it is between color spaces of same dimensionality such as RGB and HSV).
Is our computed averages and medians space invariant? Of course not: we are computing them on a per-channel basis, the base vectors coupling is absolutely not preserved.
Which of our four values should we use then? It really depends what we want to minimize; using directly a normalized HSV is indubitably a bad choice, since a 10% brightness variation impact as much as a 36 degrees of hue phase switch -- we would have to create some meaningful subspace before.
For simplicity, that's why I used RGB, since a same variation regarding one component have comparable impact when applied to others.
As for how much images are off-track compared to colors, that's not that bad; I wasn't expecting more with such a restrained sample of colors(502)... Average distance to named color is 19.53 , standard deviation 10.02 1/256th of units.
Originally posted 77 months ago.
(permalink)
syfou3 edited this topic 77 months ago.
|
|
|
Just to let you all know I am on the road right now, away from my ordinary machines, so the automatic tagging will be on hiatus for a few days (until February the 16th)... I prefer to do this this way since I am harder than usual to reach.
Posted 77 months ago.
(permalink)
|
|
|
I just came across this article about advanced tagging and it's implications:
geobloggers.blogspot.com/2006/01/advanced-tagging-and-tri...
Originally posted 76 months ago.
(permalink)
Quasimondo edited this topic 76 months ago.
|
|
|
There's one more thing I'd like to propose to improve the usefulness of colortagging: I think there should be a general "colortagged" tag added to images that have been processed, similar like geo tagged images have the "geotagged" tag. Like this one could search for those improved images - because right now due to the nature of tags you cannot search for something like "color:hsv_avg=*".
Originally posted 76 months ago.
(permalink)
Quasimondo edited this topic 76 months ago.
|
|
I agree - just like "geotagged", adding a "colortagged" field will be the best way to filter and find images in the group that have the color tags. Without it, there's no way to obtain only images that have the tags.
Posted 76 months ago.
(permalink)
|
|
|
I do not mind at all adding a tag such as "colortagged": we just have to make sure everyone else is fine with it...
But frankly, look at the API: it is programmatically simple to look up tags based on a given pattern, so this redundancy is really just needed for manual checks via the web interface; aren't most uses of those tags going to be through some kind of automaton anyway?
Originally posted 76 months ago.
(permalink)
syfou3 edited this topic 76 months ago.
|
|
|
I could easily be wrong, but I thought the api didn't allow a partial tag search. So, for example, I don't think I can call flickr.groups.pools.getPhotos and pass it "color:" or "color:*" and get back photos that have the tag "color:hsv_med=FFFFFF"
Posted 76 months ago.
(permalink)
|
|
|
As we know there are two ways you can search - tags or texts. Searchig for the "color:hsv_med" in tags obviously gives 0 results. Searching for the same in texts surprisingly returns 11 results. Still that's a bit strange I think. Why only 11? Where are the others?
So I'm curious syfou - how would you implement a pattern search programmatically which does not involve downloading all images of a group and all their tags first? I dare to claim that it is not possible - I'm pretty sure the geotaggers would have found a way already as they are doing this for a bit longer.
Here's the direct link to the API explorer photo search: www.flickr.com/services/api/explore/?method=flickr.photos...
And here for the group search: www.flickr.com/services/api/explore/?method=flickr.groups...
Posted 76 months ago.
(permalink)
|
|
|
+1 to adding a "colortagged" tag!
Originally posted 76 months ago.
(permalink)
psd (a group admin) edited this topic 76 months ago.
|
|
|
I realize what I wrote was misleading. There is no reliable way to ask directly for a group of tags based on a pattern (at least that I am aware of), but it is simple to do so by combining two types of calls: merely listing all the photos in the group, then ask for their tags, one request per photo (no need to download the images themselves); about 92 % of all the photos in Color Fields are tagged anyway... Here is a code snippet to do just that.
But I did not want to raise a whole debate here; if general opinion is that adding a "colortagged" tag is a good idea, I just will, and there is already one administrator (psd) who think it is.
Originally posted 76 months ago.
(permalink)
syfou3 edited this topic 76 months ago.
|
Would you like to comment?
Sign up for a free account, or sign in (if you're already a member).
|
Color Fields (let Paint Chips be your guide!) /
Discuss
Feed – Subscribe to Color Fields (let Paint Chips be your guide!) discussion threads