MicrostockGroup Sponsors


Author Topic: Photographic Tools  (Read 5399 times)

0 Members and 1 Guest are viewing this topic.

« on: September 04, 2015, 14:07 »
0

Guys,

I had three ideas I wanted to run by you guys for feedback on as photographers. Im thinking about building some cool machine learning products. Here are the ideas. Please let me know if these would be something you maybe interested in. Any feedback would be amazing. I think some of these tools could be pretty cool:


1. Graduation Photo Organizing Tool

Idea is if a graduation photographer has a bunch of photos, they may take a lot of time to sort them (by name). Here are some relevant links below. Use facial recognition to auto-tag them by the event.

http://www.briancbecker.com/blog/research/web-scale-face-recognition/
https://paw.princeton.edu/issues/2015/07/08/pages/5609/index.xml

2. Sports Photography Tool

Idea is if a sports photographer has a bunch of photos, organize them by player name. Relevant Links below:


http://www.gettyimages.com/editorial/new-york-yankees-pictures

http://www.corbisimages.com/Search#p=1&q=yankees+2014

http://www.apimages.com/MLB-photos

http://www.apimages.com/Search?query=baseball+and+source%3Dap&ss=10&st=kw&entitysearch=&toItem=27&orderBy=Newest

http://www.mlb.com/photos/gallery.jsp?content_id=98072634&c_id=nyy

http://newyork.yankees.mlb.com/photogallery/archive.jsp?c_id=nyy

http://mlb.mlb.com/photos/index.jsp

http://espn.go.com/mlb/team/photos/_/name/nyy



3. Auto-tagging. Automatically generate high-quality tags. I know this has been discussed a lot recently, but wanted to throw it out there still. Link below


http://www.microstockgroup.com/general-stock-discussion/someone-to-help-keywording-images/msg23765/











« Reply #1 on: September 06, 2015, 20:32 »
0
anyone? any feedback?

PZF

« Reply #2 on: September 07, 2015, 01:43 »
0
Rather specific, no? I don't need 1 or 2 and don't understand 3 when we have many tools for this already. Or we can pay somebody.
Sorry.

« Reply #3 on: September 07, 2015, 10:14 »
0
Do you ever do sports photography or graduation photography? it would seem to be a real pain to me.

« Reply #4 on: September 07, 2015, 10:32 »
+2
The problem isn't doing these things, it's doing them well enough that the time taken to clean up any mistakes isn't greater than just doing it yourself in the first place.

In the case of auto tagging, I think that's an especially tough nut to crack (doing it well).

Easy for isolated fruit & veg, but how could you possibly deal with places accurately?

Or things that any human could grasp but would be hard to get right automatically. Put a basket on a blanket in a grassy field and most people will think picnic basket. I have an image of a very messy room; it sells because of the mess (it's a teenager's messy bedroom). Two of the top keywords are messy and bedroom - how would software figure out the difference between a lot of objects tidily arranged and in a jumble, or a bedroom from another room (when the bed is hard to identify because it's messy too)?

« Reply #5 on: September 07, 2015, 15:43 »
0
I think it can be pretty good Jo Ann. Not that good, but still pretty good and useful. What about the sports or graduation idea? If they were good, would you use such products?

PZF

« Reply #6 on: September 08, 2015, 03:44 »
0
Do you ever do sports photography or graduation photography? it would seem to be a real pain to me.

No I don't. Which is the point I was trying to make - it seems rather specific.
I wonder how many/few people here do this.

50%

« Reply #7 on: September 08, 2015, 04:28 »
0
I would like to see the auto-tagging/keywording tool.

« Reply #8 on: September 09, 2015, 14:08 »
+1
Does anyone want to see any demos? Please let me know. Any any other feedback? Honestly, I'm excited about building next generation tools for photographers. I just need some feedback on whether you would use them or not.

« Reply #9 on: September 10, 2015, 08:46 »
0
I had a look at imaggas demo http://imagga.com/auto-tagging-demo# for auto-keywording.  It was impressive but useless so far as my needs are concerned.  Other than the very valid point already made about location it just seems it would take longer to correct the errors than to keyword from scratch.

It was very clever in recognising one of my test images as a statue. But it wasnt a statue, it was a metal sculpture so all the derived keywords were off on the wrong track.  Similarly, it recognised a cake as a shield  (the baker will not be happy) and again generated a whole bunch of invalid keywords, Greek, ancient etc.

I dont quite use a controlled vocabulary for stock photography as I have in other applications but I do want control over the terms so images can be considered similar on my terms not the auto-taggers so, if I use narthex, I want porch, church, Orthodox, Christian, architecture, architectural generated but how likely is an auto-tagger ever to recognise narthex in the first place?

As my son is looking at this area for his degree project I would be interested in anyones comments with a more positive view on auto-tagging.
« Last Edit: September 10, 2015, 09:06 by douglas »

« Reply #10 on: September 10, 2015, 10:07 »
0
That's really interesting @Douglas. I think some of the keywords you list maybe possible at high accuracy.

Photographers, any thoughts on graduation photos idea or sport photography photos idea?

« Reply #11 on: September 10, 2015, 11:08 »
0
If it is a commercial photo (i.e NHL) where every second is of the essence, yes facial recognition/tagging would make sense. 

Unless you have personal info for the people you are tagging it is pretty pointless.  If you are proofing online, they would appear in order and presumably they collect their diplomas alphabetically so that wouldn't be an issue.  The moms have Rebels, the kids have Iphones and the parties have instant photo booths.  The only thing I really do anymore is an actual portrait sessions and sometimes a school will ask me to come take a class photo on the steps of Parliament or at their banquet hall for example.  Sometimes I have a portrait client AND I'm at the ceremony I will shoot my clients - the parents always buy these shots.  But a whole class on spec?  Not sure it's not worth the effort any more. 

Sports - again, are you wondering about spec shooting games?  I can't see face recognition being very accurate with helmets and cages and visors.  I don't see it as necessary either unless you need to wire a newspaper.  The only guarantee of sales in action photography is to be commissioned by red #9 and shoot a game for him.  Ask him to split the fee with a couple friends and you have a good nights work.  I've gone in to shoot one kid and they are injured in the first 45 seconds. 

No, I would never use facial recognition.  If I was shooting the Olympics on the other hand.....

« Reply #12 on: September 11, 2015, 08:43 »
0
If it is a commercial photo (i.e NHL) where every second is of the essence, yes facial recognition/tagging would make sense. 

So NHL photographers is good use case?

Unless you have personal info for the people you are tagging it is pretty pointless.  If you are proofing online, they would appear in order and presumably they collect their diplomas alphabetically so that wouldn't be an issue.  The moms have Rebels, the kids have Iphones and the parties have instant photo booths.  The only thing I really do anymore is an actual portrait sessions and sometimes a school will ask me to come take a class photo on the steps of Parliament or at their banquet hall for example.  Sometimes I have a portrait client AND I'm at the ceremony I will shoot my clients - the parents always buy these shots.  But a whole class on spec?  Not sure it's not worth the effort any more. 

I mean, have you seen a SmugMug listing of a university graduation? Thousands of photos all unorganized. See what I mean?


Sports - again, are you wondering about spec shooting games?  I can't see face recognition being very accurate with helmets and cages and visors.  I don't see it as necessary either unless you need to wire a newspaper.  The only guarantee of sales in action photography is to be commissioned by red #9 and shoot a game for him.  Ask him to split the fee with a couple friends and you have a good nights work.  I've gone in to shoot one kid and they are injured in the first 45 seconds. 

Ok, for 80% or more of these photos here: http://www.gettyimages.com/photos/new-york-yankees?sort=mostpopular&excludenudity=true&mediatype=photography&phrase=new%20york%20yankees&family=editorial, you either have a jersey # or decent face. So the idea would be to automatically tag them specific to game. Make sense? How do photographers usually sort their shots when they have a few hundred? Do they manually sort them and then sent them to Getty?

No, I would never use facial recognition.  If I was shooting the Olympics on the other hand.....

So Olympic photographers is good use case? The idea is in general to create a sports database that is oriented to the specific event (NHL event/Olympic event, etc.) So the accuracy would be much higher. Make sense?

« Reply #13 on: September 11, 2015, 09:46 »
0
If you want to search for it, earlier this year Scott Kelby wrote a fabulous article about shooting NFL and the pressure of tagging and wiring those photos.  (Sorry, I can't remember if it was on his blog, Kelby One, or if it was a guest post somewhere.)  I think tagging is a much tougher job than facial recognition, but obviously when you have 7 minutes to dump a card, cull, edit, tag and wire 3 photos from 1000 every second counts.

I believe we are having two different conversations.  Would I personally use facial recognition?  No, never.  90% of my income is youth sports/school/dance.  The largest graduation class I have had is maybe 125.  What are you talking about, convocation where they hand out diplomas?  Why aren't they uploaded sequentially?  Again.... diplomas are handed out by class and usually honors/awards then alphabetical?   Why would it be so hard to lump each class in a folder sequentially?  Shooting soccer tournaments is organized into team/game folders.  Up to parents to find their own proofs in that game's folder. 

« Reply #14 on: September 11, 2015, 10:14 »
0
Let me know if you find Scott Kelbys article. I would be interested.



I'm thinking more about large scale graduation events like this one. There are a bunch of unlabeled photos. Make sense?
http://pawprinceton.photoshelter.com/gallery/Commencement-2015/G00003bGK1kafX20

« Reply #15 on: September 13, 2015, 17:49 »
0
Im assuming these are the two articles:


http://www.slrlounge.com/inside-the-nfl-scott-kelby-on-shooting-the-falcons-seahawks-football-game/


https://aepoc.wordpress.com/2012/10/19/scott-kelbys-the-challenge-of-shooting-hockey/


I'm not sure they depict use cases any better, though I do see the challenges. Anyone else have any thoughts on these tools?

50%

« Reply #16 on: September 14, 2015, 10:17 »
0
I'm still interested in auto-tagging, I think the example on imagga is very good for normal not too specialized pictures if you limit the results to the first 10 (highest percentage). Something like this that automatically tags a folder in IPTC and I would be very interested.

« Reply #17 on: September 20, 2015, 23:04 »
0
so no one really likes the facial recognition ideas?

« Reply #18 on: September 24, 2015, 18:28 »
0
No other thoughts on computer vision applied to photography?

« Reply #19 on: February 16, 2016, 04:05 »
0

That is amazing I read and check also it looks interesting to me..But we have to fulfil the keyword requirement to do it 100% accurate..and easy..

Is anyone can give me an idea that using keyword which low phrases can helo to get it advertise easy ???

« Reply #20 on: February 20, 2016, 21:00 »
0
Guys,

I have some fairly advanced solutions for both tagging and facial. Would anyone be interested in trying them? I will be able to demo them from March 14th on-wards.

Also, if anyone has further thoughts on their utility in photography use cases discussed, please let me know.

Thanks.

« Reply #21 on: February 21, 2016, 02:10 »
0
I shoot people on white. What I'd be interested in is auto tagging what the person is doing in each photo. ESPECIALLY the unique things they are doing in each photo. Say I spend an hour photographing a businesswoman on white doing various things. There will be a ton of keywords that applies to almost all images, such as, businesswoman, woman, caucasian, suit, but then there will be unique keywords that only applies to some images. For example if the woman is pointing with her hand, then pointing will be a unique keyword. If she is jumping, then jumping will be a unique keyword.

I'd be pretty impress if your software can identify the unique things about each image.

« Reply #22 on: February 23, 2016, 15:39 »
0
@Charged, Yes, can you send me some photos? I will run a few images.

Anyone else interested?

« Reply #23 on: February 26, 2016, 21:54 »
0
@Charged, Dropbox links are fine.

« Reply #24 on: March 03, 2016, 09:38 »
0
Anyone else interested in trying these solutions? Please let me know.

« Reply #25 on: March 14, 2016, 20:58 »
0
Anyone else interested in trying these solutions? Please let me know.

« Reply #26 on: May 15, 2016, 07:36 »
0
Anyone else interested in trying these solutions? Please let me know.


 

Related Topics

  Subject / Started by Replies Last post
11 Replies
4057 Views
Last post March 20, 2008, 18:10
by PenelopeB
0 Replies
1938 Views
Last post August 31, 2009, 11:30
by News Feed
24 Replies
4930 Views
Last post March 11, 2012, 08:37
by digitalexpressionimages
21 Replies
5079 Views
Last post May 28, 2014, 18:31
by shotupdave
2 Replies
1167 Views
Last post May 15, 2016, 07:36
by jsniff12

Sponsors

Microstock Poll Results

Sponsors