Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Doof]

Picasa 3.5 finally brings facial recognition to the desktop and from what I've tested with it so far, it's fantastic.

The only problem I've seen is that it doesn't appear to actually write any of this information into the file tags anywhere.

Has anybody come up with a workflow to get this information from Picasa into MC? Some kind of playlist export function or something similiar? Anything?

[Doof]

Oh, found it... All you have to do is select one of the "Person" albums in Picassa, and then Tag it with that person's name. That tag will show up in MC as a Keyword.

Sweet.

[raldo]

Oh, found it... All you have to do is select one of the "Person" albums in Picassa, and then Tag it with that person's name. That tag will show up in MC as a Keyword.

Is the tagging rectangle stored as XMP info in the photo?

[robydago]

Is the tagging rectangle stored as XMP info in the photo?

seems all info are stored in .picasa.ini files; those files are created in the same folders where the "people" pictures are.

[Doof]

Yeah, obviously MC isn't going to be able do anything with the actual face rectangle. I was mostly looking for a way to just tag my photos in MC using Picasa's facial recognition to ease the pain. I think I may just start using Picasa to manage my photo collection, but I'd still like to be able to browse my photos by person in Theater View, hence the need to get that data in there somehow.

[CadErik]

Is the tagging rectangle stored as XMP info in the photo?

At least WPG does store facial tagging information in the XMP in some MS specific fields.

Erik.

[HTPC4ME]

so doof. it actually works well?

i have hundreds if not thousands of personal pics, and if it can eliviate some headaches and speed up my taggin process i think i may try it.

[CadErik]

so doof. it actually works well?

i have hundreds if not thousands of personal pics, and if it can eliviate some headaches and speed up my taggin process i think i may try it.

WLPG does only locate faces in pictures and adds dialogs to name or drag names easily. WLPG doesn't require to upload photos unlike Picasa. In addition, since its tagging is XMP based, it can be reused in other programs.

Erik.

[Mr ChriZ]

Dumb question what's WLPG? 

[StFeder]

http://en.wikipedia.org/wiki/Windows_Live_Photo_Gallery 

[robydago]

WLPG doesn't require to upload photos unlike Picasa.

What do you mean?

AFAIK you can use face recognition in Picasa without doing any photo uploading

[Doof]

Picasa 3.5 adds the ability to do facial recognition without having to upload anything anywhere.

You import your photos, and Picasa analyzes them. It groups photos together that it thinks are the same face. You then give that group a name. If there are some photos that Picasa thinks are matches, then it flags them with a ? and gives you the option to confirm them.

It works really, really well. I tagged roughly 3000 photos last night in about an hour. It would probably have been even faster except that I didn't realize I had all kinds of DVD cover art from DVD Profiler in my Documents folder and Picasa imported all of that as well. So I had to tell it to ignore the faces of everybody it picked up from that.

Even when I got through the easy stuff and was down to the images that Picasa was fumbling on, mainly images of faces that were in the background of the photo, it was still really simple to tag those faces manually.

[HTPC4ME]

doof is there software out there that will do recognition of all photos? i have hundreds of thousands of graphic pics... be cool if there is a proggy that will find\tag pics with trees, sunsets, tigers, bears, cars etc...
would save a LOOOONG time for tagging. and make it more possible to accomplish.
TY

[Doof]

doof is there software out there that will do recognition of all photos? i have hundreds of thousands of graphic pics... be cool if there is a proggy that will find\tag pics with trees, sunsets, tigers, bears, cars etc...
would save a LOOOONG time for tagging. and make it more possible to accomplish.
TY

Sorry, I don't know of anything like that. But if you find something, let me know!

[HTPC4ME]

i been looking. cause i know my pictures will be  a nightmare. but sound slike the picaso facial will help me for the personal pics at least... ty for the tip.

[rick.ca]

It can go a step too far. I still haven't decided what to name this guy.  

[JimH]

PicFace

[raldo]

Yeah, obviously MC isn't going to be able do anything with the actual face rectangle. I was mostly looking for a way to just tag my photos in MC using Picasa's facial recognition to ease the pain.

I'll continue using Windows Live Photo Gallery (WLPG) just because of the fact that it stores the face tagging data rectangle in the image. The location of the person in the photo is valuable information, so I'll endure the marginal extra time.

By the way, what is the hit rate of guessing the correct faces in Picasa?

[galahad1974]

I've used picasa for about 150,000 images now, and its running about 93% correct, 5% incorrect but for very good reason, and 2% just dead wrong.

[raldo]

I've used picasa for about 150,000 images now, and its running about 93% correct, 5% incorrect but for very good reason, and 2% just dead wrong.

So, once a face has been detected, 93% correct? That's pretty good!

What's the rate of actually detecting a face (shadows, face rotated, profiles etc.)? I.e, how faces have you added manually out of your total faces?

[Doof]

I'll continue using Windows Live Photo Gallery (WLPG) just because of the fact that it stores the face tagging data rectangle in the image. The location of the person in the photo is valuable information, so I'll endure the marginal extra time.

By the way, what is the hit rate of guessing the correct faces in Picasa?

Picasa stores the rectangle around the face. There's just no way to get that information to MC because MC doesn't display that kind of information. There really is no place for WLPG right now, IMO. Seriously, just install Picasa 3.5 and let it do it's thing. It doesn't appear to add any information to the tags themselves so if you don't like what you see, there's no harm done. You'll get the best of both worlds with Picasa. The location of the person in the photo, and really excellent facial recognition/matching.

[raldo]

Picasa stores the rectangle around the face. There's just no way to get that information to MC because MC doesn't display that kind of information.

The main reason why its impossible to get that information into mc is that Picasa uses a proprietary database.

There really is no place for WLPG right now, IMO. Seriously, just install Picasa 3.5 and let it do it's thing. It doesn't appear to add any

Since WLPG uses XMP to store the information in the image, this info follows the picture if you decide to share the images through any channel. I made a plugin to import the WLPG information into MC.

Unlike picasa where there's a proprietary DB and you have to share using Picasa Web. Google is locking you in.

I made an experiment yesterday: Import a folder with 210 images and see how Picasa compares to the tagging already done in WLPG.

I admit, the user interface of Picasa is better and it detects more faces automatically. The recognition part is impressive although it recognized only 1 out of 8 people on the average. The remaining 7 people had to be tagged manually.

Out of the total head count, Picasa detected around 30% of the people. There are difficult cases yes: dark areas, partly covered faces.

This percentage will be very dependent of what kind of photos you have. In my case it was a vacation trip with many relatives over a 2 week period with many different types of situations.

But my point here is that depending on the accuracy you want, you must do some manual work. And since WLPG stores info in the tag, I feel that this fact outweighs the marginal extra amount of manual work in  WLPG.

[Doof]

The main reason why its impossible to get that information into mc is that Picasa uses a proprietary database.

I was specifically talking about the rectangle data. MC doesn't even show it, so it really doesn't matter how Picasa stores this data. There's no way to move it to MC. You can't move it from WLPG either.

But my point here is that depending on the accuracy you want, you must do some manual work. And since WLPG stores info in the tag, I feel that this fact outweighs the marginal extra amount of manual work in  WLPG.

The whole point of this thread was getting the people tag information out of Picasa into MC. I found Picasa's facial matching to be absolutely fantastic so I was looking for a way to leverage that ease. There will obviously be a need to do some manual tagging, but with WLPG it's 100% manual. Even if Picasa only got 30% of it for you (for me it was probably closer to 80-90%), then it's worth it, IMO. It's far easier to let Picasa do the bulk of the sorting work and then manually add a Keyword tag to each "person group" of photos in Picasa. You can then easily have MC copy that information to your People tag. Compare that to manually having to provide a name for each individual face in WLPG and then migrating the data over.

[raldo]

I was specifically talking about the rectangle data. [...] You can't move it from WLPG either.

Yes you can. Read my previous post again.

The whole point of this thread was getting the people tag information out of Picasa into MC. I found Picasa's facial matching to be absolutely fantastic so I was looking for a way to leverage that ease.

Yup, thanks for the info. Last I checked, Google Picasa only did face detection and recognition on the web app, so this is definitely a step forward.

My point was merely to gather info on how it works. I then tested it myself to compare. Hopefully both our observations will benefit other users...

It'll be interesting to see if they decide to expose the face info, preferably through xmp. I'll bet they won't because they want to lock users into their web application (cloud).

There will obviously be a need to do some manual tagging, but with WLPG it's 100% manual. Even if Picasa only got 30% of it for you (for me it was probably closer to 80-90%), then it's worth it, IMO. [...] Compare that to manually having to provide a name for each individual face in WLPG and then migrating the data over.

True, as I said earlier. And if Google decides to writes the rectangle info as XMP (like WLPG) I'll start using Picasa.

[Doof]

Yes you can. Read my previous post again.

Ok, I'll bite. You've migrated the rectangle data from WLPG to MC. Now how do you actually use this information from within MC?

[raldo]

Ok, I'll bite. You've migrated the rectangle data from WLPG to MC. Now how do you actually use this information from within MC?

Currently, it's just used to order the people correctly from left to right.

Maybe JRiver will somehow draw this info in the future, directly based on the XMP data? Then they could leave the detection/recognition to someone else, and just do the display part.

In the meantime, I'm researching how to draw on top of display view. Maybe I'll be able to get something reasonable working in the near future.

[MusicHawk]

In a recent posting in Google's Picasa forum, someone identified as Google Employee said they are looking at storing facial recognition in image tags. What this might mean isn't stated -- the names identified in the photos, or an ID that then can lookup the names in a database, or what? Also, there's no statement about "when" or even commitment to do so.

The comment was in response to questions about how to backup Picasa facial data, and especially how to share it among home PCs. What MC does so smoothly via its built-in Library Server doesn't exist at all in Picasa.

The other reason to store data/tags in the image files was to allow sending pictures to family/friends and having them see the people names because it was in embedded tags. This is totally lacking in Picasa's automatic behavior, but can be forced in a bit by manually tagging in Picasa. Of coruse, this is also only partially available in MC because it reads but does not update certain "standard" tags.

The whole thing is a huge mess due to lack of true STANDARDS to comprehensively tag images. Therefore, whatever product is used to do all the tagging labor might be the only product that then CAN be used to retrieve the tags. Scary...

BTW, Google stores facial/person data in two chunks. An .ini file in the folder with the picture has data about its face content, and IDs of names that have been matched. The names themselves are in Picasa's proprietary database on the PC used for the tagging. Nothing is put in the picture file. So, once images are processed by Picasa for facial recognition, it might be dangerous to further manipulate them using MC or anything else. For instance, if an image file is renamed and/or moved, Picasa's .ini for that image will become orphaned and the work of identifying faces in it will be lost. Image this happening to many, many image files and the risk is huge.

The only workaround, and perhaps the only safe use of Picasa, is the trick of having it recognize faces and putting them in person albums. Then do manual tagging to put the person names into the image files, then disregarding Picasa's .ini and database system.

OPPORTUNITY? Assuming it's out of JR's field to build a facial recognition system, perhaps it could be feasible to build an MC plug-in that could process Picasa data, import it into MC, then also store it in image file tags -- the crucial pieces Google left out. This would require reverse-engineering Google's Picasa database and .ini file content, but if possible it could make MC a must-have companion to Picasa for everyone who realizes how fragile and limited Google's system is.

[Doof]

The only workaround, and perhaps the only safe use of Picasa, is the trick of having it recognize faces and putting them in person albums. Then do manual tagging to put the person names into the image files, then disregarding Picasa's .ini and database system.

This is exactly what I'm doing and what I talked about in my second post.

[HiFiTubes]

This is exactly what I'm doing and what I talked about in my second post.

By tagging and then using the Keyword field right?

I just tagged about 3K files myself, and now still have about 3000 unnamed persons which seems to be lower res. images as many are persons I have already tagged/named.

Or, maybe even though the scanning is complete, it seems as though Picasa is still working through all the scanned data and sorting it.

[Doof]

By tagging and then using the Keyword field right?

I just tagged about 3K files myself, and now still have about 3000 unnamed persons which seems to be lower res. images as many are persons I have already tagged/named.

Or, maybe even though the scanning is complete, it seems as though Picasa is still working through all the scanned data and sorting it.

Yeah, I've been letting Picasa sort my photos into various people albums, and then I just select each album and add a keyword tag that matches the person's name. Then I just have MC copy the keyword tag to the people tag. I'm not using the keyword tags for anything currently, so this works out easily enough. If you were using the keyword tag, you could create a temp tag and have MC move the keyword data to that temp tag and then put it back later when you're done with the people tagging.

[HiFiTubes]

A few dumb questions:

I don't see how to "TAG" people albums in Picasa to push that data into the keyword field.

I have about 50 people  in Picasa, and assumed you meant I need to further tag them to see something show up in thie field when using MC14.

Also, Picasa let's one select multiple people and discover pictures containing a number of individuals. Are these people pushed into keyword as a semicolon delimited list?