Ger Bosma 說︰

Thanks Matthew

I made this a sticky topic.
Hopefully our group members will want to give this a try. (-;


@all: Feel free to give feedback here.
56個月前 (永久連結)

WalrusTexas 說︰

From the instructions:
"Rate each photo, where 1 is the lowest, and 5 is the highest. Try not to judge the subject matter, but instead judge based on things like composition, focus, exposure, contrast, etc. "
Having tried this out, what's your impression? Do you see significant variation in technical quality among these examples?
56個月前 (永久連結)

jamelikat 說︰

Hallelujah is my automatic response ! : )

2ndly, sure, will make some time with the scoring tool.

Th you
jAm
最初發表於 56個月前。 (永久連結)
jamelikat 編輯此主題 56個月前。

cliveg004 說︰

WalrusTexas:
Hi there, just given it a go. A fairly typical range of images I guess, some good, some not. Adding a quality algorithm should add a useful dimension to Explore - anything to get rid of lego and buses!
56個月前 (永久連結)

Ger Bosma 說︰

Art-G:

Next take out all of those who have 1000's of followers but can only be bothered to follow a handful in return - not sure what kind of ratio but there are many like this who don't give anything back.

Art-G 說︰

Ger Bosma:

I totally agree with all you have said - there are exceptions and you appear to be one of them.

I do block people who follow me but are inactive but for you it would be quite a task, I can see that.

My idea wasn't perfect admittedly, but I have seen people end up on Flickr with 1000's of Likes for their photos regardless of whether any particular photo makes it onto Explore.
56個月前 (永久連結)

WalrusTexas 說︰

Art-G:

Probably lots of things about Flickr could be improved in various ways. This thread is about one particular project that is not just brainstorming or idle speculation, but a serious attempt to enlist our help in developing a specific alternative to the current algorithm. Let's work on that!

Very much interested in everyone's thoughts about this project.
56個月前 (永久連結)

WalrusTexas 說︰

Do you find it difficult to disregard the subject while assessing technical competence? For example, I've seen very few photos that are not well exposed, but I have seen some that are very high key, with all the highlights blown out. In those cases, I judge that the "overexposure" is a consequence of the photographer's intent to emphasize the subject and set the photo apart from one that might be less dramatic or confused with a more conventional brightness histogram.

(& I reflexively downgrade dolls & legos, even when the only technical problem is boring composition, with the subject placed dead center.)
56個月前 (永久連結)

wjpostma 說︰

wjpostma:

and maybe add a weight in relation to amount of landscapes, portraits, macro's etc; fluctual (unpredictable) per day
56個月前 (永久連結)

jmunt 說︰

WalrusTexas:

Do you find it difficult to disregard the subject while assessing technical competence?

cliveg004 說︰

WalrusTexas:
Not really, once you get your eye in. I do an Admin job on a group which has all manner of content including a lot of things that don't particularly interest me - however, if the image is taken and processed well it will still stand out no matter what the subject.
最初發表於 56個月前。 (永久連結)
cliveg004 編輯此主題 56個月前。

WalrusTexas 說︰

wjpostma:

wjpostma:

apg_lucky13 說︰

WalrusTexas:
"Probably lots of things about Flickr could be improved in various ways. This thread is about one particular project that is not just brainstorming or idle speculation, but a serious attempt to enlist our help in developing a specific alternative to the current algorithm. Let's work on that!

Very much interested in everyone's thoughts about this project. "


Agreed. What can I do to help?

Jason
最初發表於 56個月前。 (永久連結)
apg_lucky13 編輯此主題 56個月前。

WalrusTexas 說︰

apg_lucky13:

Follow the link at the top of this thread, and rank some photos.
56個月前 (永久連結)

Ger Bosma 說︰

What I've noticed in general is that I rate pictures 1 or 2 that are:

- Technically subpar: not in excellent focus, too much motion blurring, badly composed, over- or underexposed (silhouettes etc), horizons not level, weird/garish colours or not crisp enough, extreme pixelation, too much sharpening or too soft, image noise, ugly lens flares that add no merit etc.

In wildlife photography (my own subject matter) I additionally rate images 1 or 2 that are too cluttered, e.g. birds obscured by a tangle of branches, too shallow depth of field, and weird or ugly crops.

- Content-wise subpar (in my subjective view): anything lego or other dolls, most forms of transport, flowers, second life (basically anything not a photo, or obviously photo-shopped), human models that are too plain/generic/boring, random geometric shapes with little esthetics, random street scenes, fence-shots, random bokeh shots, anything with drops/reflections in drops, anything with corny (romantic) symbolism (heart-shapes, padlocks, rings, hand-holding), animals that are clearly pets, miserable zoo animals in cages, city scenes without any urgency (also reflections for reflection-sake), arty-farty b/w that does not add but subtracts from the scene, sunsets for sunset-sake (brightish colours in the sky, but otherwise badly exposed and anonymous scenery).

In wildlife/bird photography: mediocre poses, images lacking detail or taken from too great a distance (2% bird, 98% surroundings) ugly or unsuitable backgrounds, humanmade perches, crappy flight or action shots etc.


NB At a later stage I will give also some feedback on the occasional 4 and (very rare) 5 star shots.


Matthew Almon Roth: I was surprised that quite a few images sported logos and signatures (-;
56個月前 (永久連結)

jamelikat 說︰

Thanks for doing this. Done 100-200 so far. I assume we can get out of there when we tire out and get back to the tool as many times as we like.
jAm
最初發表於 56個月前。 (永久連結)
jamelikat 編輯此主題 56個月前。

Karsten Gieselmann 說︰

I very much appreciate that Flickr staff is explicitiely reaching out to the members (aka "gamers" in many HF discussions) of this group to collaborate on improving Explore! 👍👍
56個月前 (永久連結)

margaretgeatches 說︰

I think it is an excellent idea to ask members to give a scoring for photographs, if this improves the functioning of Explore it can only be good. I am pleased to know that something is happening to modify the algorithm and iron out the anomalies.
I had a go at using the tool this evening and will return and evaluate some more later.

I find that I score the majority of photographs as 3 finding them acceptable but not outstanding, I tend to be harder on landscapes, perhaps because this is where my main interest lies and I consigned a fair number to 2. Anything with no obvious subject gets, poorly photographed, oversaturated, or with obvious use of HDR etc, gets 1
I scored one Lego picture, (a pet hate!), as a 3, I had to admit it was imaginative and well composed!
I skipped most of the night sky shots as I have no experience of this specialised type of photography, but did score some which included a context for the sky.
A few made it to 4, these included one or two outstanding landscapes, one or two well photographed pets, some wild life, birds and butterflies and some portraits and abstracts, and 1 train, (not at all a favourite subject, but well composed, well exposed and well lit)

5 was only awarded to a very few, possibly three photos from about an hour of rating, two were very skilfully photographed birds.

I admit that some photographs are easier for me to evaluate, I find interiors of decaying buildings less than engaging and it takes me more time to decide a scoring, I skip the paintings and drawings, as this is surely not about photography.

I didn’t see any second life screenshots, mercifully?

I hope that this input is helpful.
56個月前 (永久連結)

WalrusTexas 說︰

jamelikat:

That has been my experience. Matthew implies that the tool is not aware of which account is rating the photos.
56個月前 (永久連結)

WalrusTexas 說︰

David Marriott - Sydney:
I would also like a convenient way to cull inactive followers, but mostly because I speculated--based on nothing more than intuition--that it was a handicap for getting Explored. Now there seems to be a real alternative for the magic donkey in development. Not clear that an AI judging technical quality will be the only criterion for Explore in future, or that it will cleverly eliminate niche interests (Legos, 2nd Life, &c). We'll see.
56個月前 (永久連結)

WalrusTexas 說︰

Quick summary of my rating process:
5 Wow!
4 Lots of these, please.
3 Promising, but how about a level horizon, or restraint in post, or a better frame, or ....
2 How did that get Explored?
1 Please god, no more dolls, 2nd Life, &c, in Explore.

interesting that several of us have chosen 5 as exceptional, 4 as good, and 3 or less as needing improvement.
I do wonder how an AI trained on only 10k images can aspire to more than competent focus (which has not been lacking anyway) and rule-of-thirds composition (which would be the big improvement over the centered subjects I see in so many of the Explored samples). However, I have faith that the designers of the process are much more expert than I am.
56個月前 (永久連結)

Arthur Kantemirov 說︰

Flickr really has the most divers and complicated selection of photography, very difficult to have a system for judging which photo should get into Explore.

After 12 years of being on flickr and uploading over 2500 images, I only had one get into explore. And sometimes when seeing someone's snaps or family pics get into explore it really made me wonder how Explore works...

Totally a great Idea to have members of flickr create input for the explore selecting mechanism!
Glad this is happening!
Hope this input really shapes flickr to it's users.
56個月前 (永久連結)

David Marriott - Sydney 說︰

Appreciate the opportunity to provide some input to the tool! Should be widely promoted... perhaps the In Explore invitation tool could send the link to each explored shot owner for the last year.

Flickr (for all its faults) is still unique from my perspective - quality, groups, comments, search and longevity. A smaller group of followers/following is fine for me. Definitely better than Instagram or Facebook. I think that it could be better promoted for those who are a bit over Instagram.

I liked being able to right click through the great shots and found new people to follow.

There are a lot of images presented in the tool that have distracting logos/signatures that would have not been Explored - right? No borders though which is good

The tool seems to have been good at picking good birding shots with some excellent ones that I appreciated even though I don't shoot birds.

Far too many "record" ie tourist shots though (boring lighting, composition and subject) => 1

I found it challenging to rate a lot of the street shots and urbex though. Paintings and illustrations?

Topless dolls ????

Clearly some togs like shooting buses, trains, toys and planes but it is rare for me to see quality in them. Happy for the algorithm to include a shot per day of this type of photography as it might be interesting for some people to view.
Interesting vs quality... hmm
最初發表於 56個月前。 (永久連結)
David Marriott - Sydney 編輯此主題 56個月前。

WalrusTexas 說︰

Karsten Gieselmann:


jamelikat:

I really am going to insist that this thread stay on topic.
56個月前 (永久連結)

WalrusTexas 說︰

David Marriott - Sydney:
According to Matthew, all of the photos presented for ranking actually did show up in Explore from 2013-2018. There was a time when those who used watermarks or borders were banned, followed by a time when they were not banned but their photos were removed by staff, followed by a time when pretty much anything Safe can appear but niche interest subjects like topless dolls and trains get culled during the day.. Lots of signatures among recent Explores, although borders need to be pretty minimal to escape the censor.
The request is that we try to make a judgement on technical merit. That's clearly important, but it's not clear to me that it will--by itself--eliminate the most common complaint; i.e., that the subjects are too often of interest only to a tiny sliver of photographers. Maybe that issue can also be tackled in future projects, if it persists.
56個月前 (永久連結)

tom-quinn 說︰

I agree that working on the Explore algorithm is a good thing....and some level of transparency regarding criteria for selections would be great if possible. Completely discounting subject matter seems inappropriate given that many great photos are great simply because someone captured a compelling image that others wouldn't have given a second thought to (I love Dorothea Lange's quote: "a camera is tool for learning how to see without a camera"). Also, my son is a great photographer....he sometimes plays with focus and exposure in ways that I certainly would not and which some might consider "poor" focus or exposure. Yet he can (and does) create very compelling and artistic images. Which raises the question of whether there is age/generational bias in judging "quality" in photography. Or perhaps age is irrelevant and it is simply ultimately a matter of personal taste.....which no computer algorithm is going to adequately assess.
56個月前 (永久連結)

tom-quinn 說︰

Ok...after posting my previous note I just perused the Explore photo gallery selections which I must admit I rarely do. So, it's pretty obvious that it is broken and right now of little value. The number of incredibly lame photos is overwhelming.....to the point that any small pride I had when one of my photos was selected is gone forever. (I'll probably get over it.)
56個月前 (永久連結)

Kat-Renee Kittel 說︰

What I want to know is why did my photo get invited to this group, a photo I took yesterday and the photo isn't even in the Explore... So it was there and now it's not... that part needs fixed. Either be in or out... not wishy washy.

I did look at the photos for today and alot are really lame. Makes me want to start a best of today group and pick out good stuff myself from each day!!!
56個月前 (永久連結)

Finn Frode (DK) 說︰

Kat-Renee Kittel:

Your questions are answered in other threads in this forum, but you can see all your Explores in the Scout tool at bighugelabs.com/scout.php?username=20262597@N08&combi... - and since your picture is there, it must have been Explored at some point. It could however have dropped out by now...
最初發表於 56個月前。 (永久連結)
Finn Frode (DK) 編輯此主題 56個月前。

jamelikat 說︰

WalrusTexas:

(oops apologies if off topic earlier)

Explore-improvement-wise, i just have the nagging feeling that whatever needs to be coded into the algorithm was already all there. Just needs to be tweaked, or added to e.g. an algorithm/AI thing that would detect frames/signatures/watermarks if they are banned factors.
The added nagging feeling i have is that the MAIN thing/reason why it's broken is lack of human input (too many silly planes/trains/automobiles/2nd life/toys that must have got enough faves perhaps legitimately, BUT ultimately is so obviously of sub-par or average quality). It is the paid or volunteer real-person final curators/cullers that need to be properly in place. AI for the moment, is the junior partner in the solution. The major part is the human photographers' eye input.. Enact that and Explore will be closer to its potential for interestingness/quality. I don't accept 'interestingness'= basically networking efforts. That is misuse of language.
jAm
最初發表於 56個月前。 (永久連結)
jamelikat 編輯此主題 56個月前。

Bill VanderMolen 說︰

I think ignoring the subject completely in this rating is not really a good idea. I have always given merit to photos I see on Flickr when the subject is really hard to capture in a photo.

If a photographer is out in the field in a difficult situation to capture a hard-to-get subject, AND has done it with great exposure, composition, etc. that to me should rate higher than someone taking a photo of a tree ornament with great exposure, composition, etc.
56個月前 (永久連結)

Francis Mansell 說︰

An issue that resonates for me when rating some of these photos - and in flickr generally - is the quality of the upload. By which I mean, quite a number of photographers, some of whom are clearly talented, don't upload photos in high enough resolution for me to "enjoy" their work. We all know why people do this and it's easy and right to be sympathetic to people trying to avoid their work being stolen - but I can't give what's obviously a good photo a 4 or 5 rating if it's in visibly low resolution and I can't see it properly at any magnification. The recent (and to me very welcome) roll-out of one-click full magnification only makes this worse. I'd say don't use this rating tool on a phone - use a decent sized tablet at least.
56個月前 (永久連結)

Clickrbee 說︰

How long will the score page be available? Is there a deadline for scoring the images?

Also, I'm noticing a trend in the photos available for scoring. Most are close-ups of birds or subjects dead center of the image. I'm noticing a minimal representation/absence of the following types of images:
minimalism
abstracts
black white
street photography
patterns
landscape
seascapes

It makes me wonder if the results of this scoring exercise are going to end up skewed towards close-ups of wildlife (animals, birds, insects) and trains due to the abundance of these types of images in this scoring exercise.

In the just the two months I've been on Flickr I've seen many very talented photographers and amazing images of all types and moods. Why are they not making it into the explore? (rhetorical) I can see why folks are frustrated.

Having to dig deep into these images to try to find a variety of compositions (leading lines, rules of thirds, etc.)
最初發表於 56個月前。 (永久連結)
Clickrbee 編輯此主題 56個月前。

WalrusTexas 說︰

Thru My Shutter:

Duration: Matthew did not state a time deadline for the project. Clearly a pool of 10k images will need more than 10k ratings, but I'm not expert enough to know what's acceptable. Presumably the project can continue until predictions stabilize, assuming the participation rate is enough to make it happen in a reasonable time.

Content: I'm guessing that each person sees a random selection. I don't know what the prior probabilities are; i.e., what fractions of the sample show birds, landscapes, &c. It's quite possible that one could see a string of similar subjects just by chance. That said, I also see a lot of centered subjects, which are otherwise pretty good. If the photographers provide them and the raters accept them, then centered will become the standard.
56個月前 (永久連結)

WalrusTexas 說︰

jamelikat:
Human input is too expensive for a site like flickr that gets a lot of uploads every day. Some problems--like the 25 Explores for one photographer today--could easily be avoided by an algorithm, but I can understand that Smugmug wants to allocate resources to big exciting changes like AI selection. In fact, It seems obvious that there is some human intervention on the relatively small pool--500 or less--that shows up in Explore every day.
56個月前 (永久連結)

Francis Mansell 說︰

WalrusTexas:

I wonder what the policy is re promoting it to members. On the one hand as you say it will need a substantial number of ratings, on the other hand (surely the $64,000 question) who do they want to be doing it? A random selection of members? A self-selecting subset of members who a) know about it and b) can be bothered? (probably the reality whether that's what they want or not) Or only people who've had a photo Explored? Either way, let's hope it ends up working better than the old tagging algorithm. I still fondly remember posting a picture of a dog, tagged as "dog", getting autotagged with "cat".

Re content: re-reading the OP, it conspicuously doesn't say that the overall selection of photos for assessment is random, only the order in which they are shown to viewers. I suspect they may actually have been chosen for this purpose quite carefully - the proportion of really excellent photos seems to be lower than in a typical day's Explore, as does the proportion of really terrible ones. And I've seen no buses so far ...
56個月前 (永久連結)

WalrusTexas 說︰

Francis Mansell:
Matthew asked the admins in this group if we would host the announcement of the project—which we were glad to do—but didn’t go into any detail on the plan, beyond what you’ve seen. Clearly it will be self-selected, and starting here guarantees at least an interest in Explore.

Pretty sure I was banned from Explore for 20 months for removing an incorrect autotag. The photo in question was dropped mid-day from #31. Still crabby about that.

I’ve seen legos, dolls, and 2nd Life images in the test set. Maybe the Bus Invasion started after the test set.
56個月前 (永久連結)

WalrusTexas 說︰

Has your ranking activity in the tool affected your photography? Have you reconsidered your personal criteria for what makes a good image? Any changes in technique or subject matter?
最初發表於 56個月前。 (永久連結)
WalrusTexas（群組管理員）已於 56個月前編輯此主題。

Clickrbee 說︰

WalrusTexas:
I've been scoring for a couple of hours each day until I can longer look at the trains and birds. lol... so many of them, over and over. :) I'm focusing on aesthetic and composition more than sharpness, saturation etc. Most of these images look like snapshots, many of them of good quality, but lacking interest or have the subject dead center in most of the images. I know I've seen many interesting snapshots with interesting compositions, but don't recall seeing any of them in the Explore page either. It just seems SO random.

It would be great if they would use some of the images they've curated in the Flickr Gallery to "teach" the algorithm about aesthetics. I'm making some galleries in my stream of images I like, to see if they help me learn what I like and help me move in that direction in my photography. Sadly, many of these amazing photographers have moved on... I wonder if they could feed some of those to the algorithm just to mix things up a bit.

Personally, I am noticing that I'm seeing how mood, creativity set an image above many others. I'm getting a notion of what not to shoot. I've only been back in Flickr for a couple of months but already I'm really getting psyched about trying new things and challenging myself. So, yes, it has helped.
56個月前 (永久連結)

WalrusTexas 說︰

Thru My Shutter:
Glad to hear it! & thanks for your contribution!

I find that the technique is an indication of intent, especially a non-standard technique. Does the technique make the photo appealing? Not always, but often an original idea is its own reward. Some photos need to be sharp almost everywhere, some mostly nowhere, most compactly on the subject. Maybe could manage that with a rule-based system rather than a neural net. Never found it much of a problem in Explore, but that means I have to hold myself to a high standard. Similarly exposure is mostly ok, if you give photographers credit for doing what they intend; e.g. gloomy or excited or objective, or ...
Saturation and contrast are harder to standardize. How can an algorithm make a good choice unless it recognizes the subject? Maybe extreme, moderate and minimal versions of these variables will be approved in proportion to their history, without regard to matching the subject. We may be in for some surprises.
Composition is of great concern to me, because there is such a strong tendency to take the shot with the thing of interest smack dab in the center, and never consider an alternative. That generates a lot of suboptimal images. Centering works well sometimes but "well enough" seems to be over-represented.
And of course, we humans all want to see a place or thing that we have never imagined, or have always longed for, something completely different. "More like this, please" can never select for such things.
56個月前 (永久連結)

Clickrbee 說︰

I'm assigning 5 to those that have great composition, clarity, saturation; 4s to ones that are just a fall just a bit short. Most of the rest are between 3s and 2s and I reserve 1 for those where I say, "what the heck?!" lol. I do take into consideration whether its a portrait, still life, abstract, etc when looking at the composition... Looking for leading lines, rule of 3rds, and whether it is overall pleasing to the eye. I also look for images where the photographer was able to create an image that is compelling, creative, conceptual, moody, etc. There aren't that many so I'm really happy when I stumble on a gem. Lately, I find that sometimes the tools gives me a long string of really great captures, then reverts to the usual trains, etc. :)
56個月前 (永久連結)

Squirrel Girl cbk 說︰

Thanks for trying to improve "explore." Today I counted 8, count them 8 pictures of garbage trucks by one person.
56個月前 (永久連結)

Clickrbee 說︰

Squirrel Girl cbk:

So frustrating.... I noticed that sometimes the buses appear earlier in the day and then a lot fall off...but still, they really shouldn't be explored. I've seen SO MANY great images by people I follow or see in groups and it blows my mind that they don't make it in. When I first signed up with Flickr in 2007 I had 1 image explored, then it dropped off and never again. I wonder what the algorithm is looking for... I hardly ever look at Explore and when I do, I fave the very best in hopes that it somehow feeds back into the algorithm. For a couple of days I saw a bit more variety in the types of images, then it reverts back to buses and other repetitive subjects. So weird!
56個月前 (永久連結)

Francis Mansell 說︰

WalrusTexas:
The more I look at this, the more I wonder how on earth an AI as currently understood could possibly learn to recognise artistically good photos, as opposed to conventionally technically competent ones that obviously take into account the law of thirds or golden ratio.

Good technique may be well disguised - at least from the point of view of an AI - for instance in the example you give of a photo with very little sharpness but what there is is exactly where it needs to be. If the "training" of the algorithm leads it to recognise quite impressionistic photos with certain qualities as being good, then it seems very safe to assume that when unleashed on all the photos uploaded in a given day it will use those criteria to choose some very improbable candidates for Explore because it won't and can't have the faintest clue what the je-ne-sais-quoi factor is that makes such photos attractive and perhaps even meaningful to some humans.

I can also see why, say, wildlife photos and other areas where simple accurate depiction is the main quality criterion might be at an advantage here - by definition a good photo of a bird is going to be sharp, with the sharpest focus usually on the eyes. Two things I've been marking photos down on are over-sharpening and its frequent bedfellow, heavily amplified noise in the image - I tend to go to the actual photo to check on this because you mostly can't tell at the size they're displayed in the training tool. I do hope the algorithm will be able to spot the artifacts of over-sharpening.

Bottom line is, I don't think a truly good / useful Explore is possible in the foreseeable future without human intervention / winnowing, due to the subjective / artistic factor. Whatever algorithm they do come up with should always be in training because it will never be anywhere near perfect - the proof of the pudding will be in reducing the number of images the human(s) required to finalise the choice have to scroll through to find 500 images worth thousands of us looking at to a manageable level.
56個月前 (永久連結)

Squirrel Girl cbk 說︰

I have no problem with less than the best images being scattered into Explore. No problem, to me, that newbies get a little encouragement with their photography. But I really hate the Lego shots and the CGI images. I can't imagine why someone would want to share pictures of garbage trucks, but even that I could deal with ONE image being posted/explored on an infrequent occasion. I do get very frustrated with trains, planes, and buses being Explored so often. In the scoring thing, I rated a train coming out of a tunnel well above 1 star because it was composed and had some art to it. All the other trains, I gave them just 1 star.
56個月前 (永久連結)

Clickrbee 說︰

I hear you! Now there is around 8 or 9 snapshots of a sports car by the same person... also some garbage trucks, buses etc.
56個月前 (永久連結)

dave dube' 說︰

Yikes, I got so involved with the comments, I almost forgot my mission...

Thanks to Leadership for keeping thoughts on track!
56個月前 (永久連結)

WalrusTexas 說︰

Francis Mansell:


Squirrel Girl cbk:

Clearly Flickr staff have a lot on their plate, and have until recently been challenged mostly by performance issues. Now they apparently feel like they have that under control, and want to do something big that will engage the customers. This project—especially using us to train the AI—feels like that. I don’t expect that they will at some point exclusively use the new AI, but rather will combine it with other tools. We’ll see.
That said, I also have the feeling that technical competence is not a big issue with Explore, or with Flickr in general, except maybe for the over-use of centered composition. The most offensive thing is multiple Explores per day for individuals with niche interests—buses, 2nd life, &c. Seems like this would be easy to fix. Most days such nonsense declines after a few hours, which suggests that staff time is being wasted. How hard would it be to restrict everyone to one Explore per day?
There are startling examples of AI competence (chess, 6-player Texas Hold ‘em, ...), so I don’t reject out of hand the possibility that a bot could select jaw-dropping, thought-provoking images from the daily flow. I do have some concern that training on the Explore catalog will underweight innovation. But if the enterprise survives, there will still be good minds behind good eyes posting in groups that tilt the scales in various directions. Maybe a cool AI and the kind of thoughtful commentary we’ve seen in this thread will be steps in the right direction.
56個月前 (永久連結)

Francis Mansell 說︰

WalrusTexas:

I completely agree re multiple images from the same account - I really do not understand how that was ever allowed to happen, and since it was, how preventing it wasn't seen as desirable low-hanging fruit for the programmers. One of my biggest hopes for this project (because I suspect sorting out the above is a given if they're starting from scratch) is that images from 2nd Life and other computer games are simply not chosen for Explore. I guess it depends what proportion of people rating photos with this tool find them really irritating - they're a guaranteed 1 from me but then lots of people are gamers and probably have a totally different attitude.

But the examples of AI excellence that you give are all about making choices using a finite set of rules - there may be vast numbers of possibilities but there's a logic involved and the success of AIs in those applications basically depends on brute force number crushing of all the options to get the best result. Making aesthetic decisions - the criteria are totally irrelevant - seems to me to be another thing altogether, because there is no logic to it. It will happen, but at the moment I think it's still a very big ask. When someone succeeds in doing something like this well, we'll know that AIs with their own consciousness are not far off, because it will be succeeding in understanding human emotions, not just at doing vast numbers of calculations.

Anyone who's been on flickr more than two or three years will remember the appalling auto-tagging algorithm, which was useless in at least three different ways: 1. inaccuracy; 2. ignoring tags that had already been applied by the uploader, compounding the former - e.g. in my experience a dog, tagged on upload as "dog", autotagged as "cat"; 3. adding utterly pointless tags like "outdoor", which could be anything from a landscape to a bus to a football game. I'm sure someone at old flickr thought this could be monetised in some way ... and that it wasn't.

One of the things that the new algorithm will need to be able to do is what the old autotagging algorithm failed at far too often: identifying content. Otherwise, lots of bus pics - albeit by different people - could get in on the same day. Or whatever. Because lots of people don't tag their photos. Maybe the new algorithm should pay attention to tags - maybe flickr could even encourage tagging by making it harder to get Explored without tags. It may already for all I know.

So I'm very mindful of the huge mountain the creators of this new AI need to climb because they need to make it much better than both old Explore and the autotagging algorithm. I wish them luck and inspiration.
56個月前 (永久連結)

Squirrel Girl cbk 說︰

Hah! I'd forgotten about the Flickr auto-tagging. It was hilarious! And a pain to delete all the tags. I remember one shot of mine was of an archaeological dwelling with a ladder leading to it, and Flickr tagged it as "giraffe"!!!!! :-D

I didn't realize that the CGI images were from 2nd Life. I'd think they could/should get all their following/kudos from 2nd Life and not from Flickr. My 2 cents on that.

I'm all for having tags be a part of the selection for Explore. I always tag my images. OK, I don't put much in for petroglyphs because I don't want to encourage visitation and potential vandalism. But I pretty much always tag something about the image, and sometimes a lot. I suppose some people could tag their photos randomly to try and encourage getting an Explore, but then maybe Flickr could have some kind of reporting system if that happened, and the poster could get blackballed or something.
56個月前 (永久連結)

John Edward 說︰

Im in.
56個月前 (永久連結)

B.E.K. Photography 說︰

Very happy to see this topic "finally" come up.

As a long time Flickr user who recently cancelled their pro account. I find it encouraging that some real effort is being made in an attempt to improve the "photo quality" component of the Explore algorithm.

This action is making me reconsider my decision to cancel Pro, the current discount helps as well :)

Cheers to the Flickr team for taking this on, I look forward to seeing a more impressive Explore tab!
56個月前 (永久連結)

David Marriott - Sydney 說︰

Can't understand why the Explore algorithm can't be an iterative improvement rather than a big bang AI implementation. Clearly simple changes can be made now (see above suggestions) and then refinements over time. Happy to include all types of photography but perhaps in roughly % terms ie if 2nd life = 2% of all uploads that day then 10 shots out of 500. If less than 0.1% then none. Same for buses/toys etc. The algorithm can identify based on tags/groups.
56個月前 (永久連結)

Matthew Almon Roth 說︰

Hi all,
I wanted to jump in with a status update on the scoring tool work.

TL;DR: You are awesome!

We've had more than 100,000 scores come in, which gives us a tremendous data set to work with. Of those scores, there were 75 photos with the exact same score from everyone, which is pretty interesting. Beyond the above segment, we saw quite a bit of agreement on a number of other photos, which we found quite useful in establishing baselines. While beauty is in the eye of the beholder, there's definitely a lot of overlap among the beholders :)

We're super happy with how many of you participated and appreciate the time you took. This will go into the overall improvement of Explore. We're simultaneously working on a number of other inputs to the algorithm and may be back with further questions or ways to get involved.

Feel free to continue to score photos with the tool if you like -- we'll continue to collect data and add it to the overall findings.

Cheers,
Matthew Roth
Flickr Team
55個月前 (永久連結)

Matthew Almon Roth 說︰

David Marriott - Sydney:

Can't understand why the Explore algorithm can't be an iterative improvement rather than a big bang AI implementation. Clearly simple changes can be made now (see above suggestions) and then refinements over time

WalrusTexas 說︰

Matthew Almon Roth:
Thanks for the update! Very glad to hear that there has been substantial participation, and that there are ongoing efforts to test changes.
55個月前 (永久連結)

Clickrbee 說︰

Thank you Matthew! Thank you for the update, it is very encouraging... just glad we have some input.
55個月前 (永久連結)

Finn Frode (DK) 說︰

Matthew Almon Roth:

Thanks for the update. Keep up the good work. :)
55個月前 (永久連結)

jamelikat 說︰

Matthew Almon Roth:

glad to hear some hypotheses/tweaks to Explore will be tested soon. Th you for the update.

jamelikat
55個月前 (永久連結)

WalrusTexas 說︰

The all seeing i:

I agree that portrait orientation is disadvantaged in flickr mosaics (and advantaged in discussion threads), but I’m restricting this thread to the AI project. Feel free to post your comment in a new thread.
最初發表於 55個月前。 (永久連結)
WalrusTexas（群組管理員）已於 55個月前編輯此主題。

WalrusTexas 說︰

fotowayahead:
dave dube':
fotowayahead:
dave dube':

All of these comments belong in the Explore Frustration thread, not this thread on the AI Project.
55個月前 (永久連結)

WalrusTexas 說︰

Philippe Renauld:
Squirrel Girl cbk:

Please post Explore Frustration on that thread.

www.flickr.com/groups/inexplore/discuss/72157668110273703/
最初發表於 54個月前。 (永久連結)
Ger Bosma（群組管理員）已於 54個月前編輯此主題。

dave dube' 說︰

WalrusTexas:

Curious, how is posting comments that might seem like 'frustration' not helping to improve Explore?

Who's the judge?
54個月前 (永久連結)

Ger Bosma 說︰

dave dube':

Sheriff Walrus and yours truly lay down the law in these parts, stranger. (-;

We try to avoid unnecessary clutter because this thread is aimed at (and started by) flickr staff.

NB Your question (and this response) also are superfluous.
Don't continue to vent your frustration here.
54個月前 (永久連結)

reinaroundtheglobe 說︰

As someone pointed me out that he didn't see any 5 star worth image in this pole and to be honest, I agree with him (although I rated just a view 5 star because there weren't any better ones).

5 star images are like these www.flickr.com/photos/albertdros and similar quality

So would it be possible to put some more high quality photo's into the pool?
54個月前 (永久連結)

reinaroundtheglobe 說︰

Matthew Almon Roth:

Is there any time frame to get this job done? I've extended my Pro membership for another year (to give Flickr a little extra time to improve after the transition). Smugmug promised us Pro-users more chance to get Explored but the opposite happened last year....
54個月前 (永久連結)

Kat-Renee Kittel 說︰

Matthew Almon Roth:

We've had more than 100,000 scores come in, which gives us a tremendous data set to work with. Of those scores, there were 75 photos with the exact same score from everyone, which is pretty interesting. Beyond the above segment, we saw quite a bit of agreement on a number of other photos, which we found quite useful in establishing baselines. While beauty is in the eye of the beholder, there's definitely a lot of overlap among the beholders :)

tom-quinn 說︰

I weighed in on this subject a few months ago and I rarely view the photos in Explore. Today I decided to take a quick look....wow are they consistently bad images. My conclusion is that it is broken beyond repair and discredits Flickr, actually making the "brand" less desirable and less credible.
54個月前 (永久連結)

jamelikat 說︰

they (Flickr) must be thinking it is not totally broken...
tom-quinn:

i just hope they announce some progress soon

jAm
最初發表於 54個月前。 (永久連結)
jamelikat 編輯此主題 54個月前。

Ger Bosma 說︰

tom-quinn:

Today I decided to take a quick look....wow are they consistently bad images.

reinaroundtheglobe 說︰

Are we done with it?

503 Service Temporarily Unavailable for 2 days now
54個月前 (永久連結)

Ian Hearn Photography 說︰

I can't access the link...
54個月前 (永久連結)

WalrusTexas 說︰

reinaroundtheglobe:
Ian Hearn Photography:

I don’t have inside knowledge, but it may be that Flickr considers the experiment complete, given Matthew Roth’s happy reaction, above. Presumably 100k responses give them enough to work with. Recognize that Flickr has different metrics than we do. If Lego customizers, Bus Spotters, and NatGeo Alumni add pro accounts faster than the rest of us, that may be enough. Lots of good bird photography still on Explore, but YouPic has a better version for those of us who just want to see great images. Photo archiving and relationship maintenance will keep me active on Flickr, despite the expectation that I will get very few Explores in future, and never on photos I consider my best.
最初發表於 54個月前。 (永久連結)
WalrusTexas（群組管理員）已於 54個月前編輯此主題。

margaretgeatches 說︰

Matthew Almon Roth:

I have been impressed by the improvement in explore over the last few days, I participated in the scoring system almost daily while it was available, as did many others. Thank you, Matthew for the work put in and the feedback. It looks as if we are now seeing the very welcome results.
54個月前 (永久連結)

Jane Statham 說︰

Well that's great. I've just tried the link and it's not working.
54個月前 (永久連結)

Padmacara 說︰

Jane Statham - spuddie7:
Still not working today, 15/03/2020...
53個月前 (永久連結)

Jane Statham 說︰

Padmacara:
Most helpful !!
53個月前 (永久連結)

Travis Estell 說︰

Based on this post in the Help Forum, Flickr is already rolling out some changes to Explore.

www.flickr.com/help/forum/en-us/72157713472686717/
53個月前 (永久連結)

Matthew Almon Roth 說︰

Ok team, we're now in the testing phase for a new algorithm behind Explore.
Blog post:
blog.flickr.net/en/2020/03/19/congrats-explore-its-about-...
Help Forum post:
www.flickr.com/help/forum/en-us/72157713543129311/

One of the weird things for this group is that we will be presenting 50% of visitors and Flickr members with the old version, 50% with the new version. So there will be two different Explores running simultaneously. We are doing this so we can track the impact of the changes with real data about how you all interact with the different versions.

Apologies to David Kracht and the Admins here for what it might do to your invitation work to the group. On the plus side (for David at least), I was pleased to see that his amazing Himalayas photo from March 16th was the top photo in Explore v2 that day. Alas, we were only showing v2 to staff at the time, so he didn't get the bump he would have from being the top photo on Explore running for the public. Screenshot here for posterity:
www.flickr.com/gp/matthewalmonroth/W9055T

Happy to get feedback or questions, either here, at the Help Forum or via FlickrMail.

thanks,
Matthew
(Flickr Product Team)
53個月前 (永久連結)

Stickwork-Steve 說︰

Hi folks, I really do enjoy looking through the explore images every day, lots of wonderful photos by all types photographers. However, how many pointless uninteresting pictures of garbage tracks and schools buses to we really need? Case in point, today's (March 18/20) explore batch features 6+ pictures of a garbage truck, why? Forgive my ignorance of the system, but can the system not weed these out? Isn't there anyone acting as an editor of what the system picks out for explore?
53個月前 (永久連結)

WalrusTexas 說︰

Stickwork-Steve:
Read the post above yours.
53個月前 (永久連結)

WalrusTexas 說︰

Matthew Almon Roth:

Matthew,
Thanks! I guess. I find this somewhat obscure. Looks like I got lucky--no school buses or 2nd Life today, in my version of Explore.

If I understand--which is not at all certain--Flickr has offered half of us different Explore pages, and intends in future to use the version of Explore that gets the best response. So if you like what you see and want more of it, fave like crazy. If you don't, snap it shut.

Best of luck!
I think I got the good one today:

53個月前 (永久連結)

WalrusTexas 說︰

Matthew Almon Roth:

Matthew,
This group always invited all the photos in Explore by using David's bot. Occasionally when things broke down, some of us would invite missing photos on request. When things went really wrong for some time, we withdrew that service. Lately, the bot has been pretty healthy.

However, looking at the Explore I am offered, very late in the day after the usual shuffles, with the top ranks having amassed multiples of 10k views and hundreds of faves, I notice that they are in none, a few, or many groups, but very few have been invited to our "in explore" pool. Scout has found some, marks some "dropped" that I still see. Don't understand how that happens.

I'll invite some by hand for the sheer joy of seeing the change, but we aren't going to make a habit of it.
最初發表於 53個月前。 (永久連結)
WalrusTexas（群組管理員）已於 53個月前編輯此主題。

reinaroundtheglobe 說︰

Lucky you! I'm on the old version today 😭 2trains in the top 10 ranking
53個月前 (永久連結)

Francis Mansell 說︰

Matthew Almon Roth:
That looks way better - I'm guessing I've got the new version. Really only two small issues: two (very good) photos by the same account near the top, and an undistinguished snap of a police car near the bottom. No buses, trains or Second Life at all! This is a huge improvement and I will be looking at Explore a lot more often as a result.
53個月前 (永久連結)

Stickwork-Steve 說︰

Yikes, I guess I have the old version as well....all kinds of buses, garbage trucks and that second life stuff.
53個月前 (永久連結)

WalrusTexas 說︰

Stickwork-Steve:
Sorry to hear it, but take heart—help is on the way! Just close it down and check back from time to time until you get the good version.
53個月前 (永久連結)

David Kracht 說︰

Matthew Almon Roth:

Apologies to David Kracht

reinaroundtheglobe 說︰

Matthew Almon Roth:

One of the weird things for this group is that we will be presenting 50% of visitors and Flickr members with the old version, 50% with the new version. So there will be two different Explores running simultaneously. We are doing this so we can track the impact of the changes with real data about how you all interact with the different versions.

Ger Bosma 說︰

reinaroundtheglobe:

Try logged out (or surfing in anonymous mode) and refresh a couple of times.
53個月前 (永久連結)

Nobusuma 說︰

I see the new version and it's way better than the old one. Images are technically superior and interesting for the most part. I'm happy not to see so many random snapshots, or buses, trains, etc.
53個月前 (永久連結)

reinaroundtheglobe 說︰

Ger Bosma:

Try logged out (or surfing in anonymous mode) and refresh a couple of times

George Plakides 說︰

Over five years since my last photo was in explore, I was informed this morning from friends' comments that my latest upload was in explore. The views generated would seem to support that but I have not been able to see my photo for myself. Is there something else I should be doing other than scrolling the Explore screen?
53個月前 (永久連結)

George Plakides 說︰

Just read some of the posts above which may explain my query.
53個月前 (永久連結)

David Kracht 說︰

Matthew Almon Roth:

One of the weird things for this group is that we will be presenting 50% of visitors and Flickr members with the old version, 50% with the new version.

Paul B0udreau 說︰

Would it be fair to guess then that, if I am seeing a rapid rise in views and favs, plus an invite from David to In Explore but, I don't see my image in the Explore page I am looking at, that I am still experiencing the old version?
53個月前 (永久連結)

David Kracht 說︰

This is what I would expect...
53個月前 (永久連結)

David Kracht 說︰

David Kracht:
but, you might think about a native flickr indicator to show the/a ranking/selection in expore on the memeber's elements.
53個月前 (永久連結)

Ageeth van Geest 說︰

Ageeth van Geest Thanks a lot! Is Maria Gotfrifda right ( Smile on Saturday) . Greetings Ageeth
53個月前 (永久連結)

suttonedward242 說︰

Hi all,
Flickr, to me, is a complex mechanism, and as such, the most complex machines are reliant on the smallest components to function properly.
that is to say, from the renowned to the unknown person, we each can play a role, weather as, in my case, i want to learn, or in the case of a seasoned pro, having a lot to offer in knowledge and experience,
we each have our own "eye" and interests, i try not to rule anything in or out, and appraise initially, from that which is easy on the eye, to that which has taken effort to achieve, and of course all in between.
The Flickr community would be nothing without it's participants, .....
53個月前 (永久連結)