It's sort of like Minority Report, I guess.

August 23, 2008 9:27 AM

the first full release of Photosynth. Stonehenge. Venice. Previously 1,2,3. PC only.
posted by signal (65 comments total)

PC only.

Well, it is Micro$oft.
posted by dash_slot- at 9:39 AM on August 23, 2008

PC only.

Well, fuck that.
posted by interrobang at 9:45 AM on August 23, 2008

Didn't work for me in IE7 or Firefox. This is after installing the program, allowing the activex control and flash player. Meh.
posted by dash_slot- at 9:49 AM on August 23, 2008

FFS. Half your links are to previous FPPs. Four posts is a bit much for some single platform plugin that breaks the internet, don't you think?
posted by ryanrs at 9:54 AM on August 23, 2008 [1 favorite]

I think this is awesome.
posted by Lord_Pall at 9:56 AM on August 23, 2008

Really? How can you tell?
posted by ryanrs at 9:59 AM on August 23, 2008 [1 favorite]

Oh well...

ScottF, Official Rep, replied 2 days ago

I've noticed that you can't get back to the install page (where we discuss system requirements) after installing - we'll look into that!

Photosynth runs on Windows XP SP2 and SP3, and Vista, and requires IE7 or Firefox 2 or 3. There are a variety of video cards out there and we haven't tested them all, but the general requirement is that you have a minimum of 32mb of video RAM, with 64MB or more recommended. DirectX7 is recommended. You should have at least 1gb of system RAM and a 2.0GHz CPU to get the most out of Photosynth. You'll also need a broadband network connection. On the Mac, Bootcamp will work, but not Parallels

That rules out thislil ole laptop, then. :/
posted by dash_slot- at 10:03 AM on August 23, 2008

No only is it Windows only, it assumes that Linux is MacOS and assures me that a Mac version is coming soon...
posted by jjb at 10:08 AM on August 23, 2008 [2 favorites]

I love how the page tells me "Unfortunately, we're not cool enough to run on your OS yet. We really wish we had a version of Photosynth that worked cross platform, but for now it only runs on Windows."

Heh. Windows 2000 for teh WIN!! (Has it really been almost 10 years? OMGWTF?)
posted by loquacious at 10:09 AM on August 23, 2008

I can't find GarageBand for Windows, dammit.
posted by bz at 10:16 AM on August 23, 2008 [1 favorite]

I'm trying to figure out what the hell this is, and the website isn't helping me.

Imagine being able to share the places and things you love using the cinematic quality of a movie, the control of a video game, and the mind-blowing detail of the real world. ... Photosynth is a potent mixture of two independent breakthroughs: the ability to reconstruct the scene or object from a bunch of flat photographs, and the technology to bring that experience to virtually anyone over the Internet. What you see on this site is the first of many versions of Photosynth. Call it beta, call it 1.0, call it whatever you want…

I think it's a photostitching app. OK, cool, but I'm missing out on the "mind-blowing" and "potent" parts.
posted by Nelson at 10:22 AM on August 23, 2008

I can't find GarageBand for Windows, dammit.

Not an especially apt comparison; GarageBand isn't a browser plugin and they're not trying to encourage people to put GarageBand-specific content on the web.
posted by George_Spiggott at 10:24 AM on August 23, 2008 [1 favorite]

For the record this runs fine under VMWare Fusion, contrary to what Microsoft seems to be saying. Venice is pretty impressive but somehow I don't see these things taking over the web.
posted by Space Coyote at 10:31 AM on August 23, 2008

For those of you it doesn't work for (everyone except for 5 people), here's a youtube video demonstration, and description. You can find a bunch more by just digging around YouTube, of course.
posted by taursir at 10:34 AM on August 23, 2008 [1 favorite]

Unfortunately, we're not cool enough to run on your OS yet. We really wish we had a version of Photosynth that worked cross platform, but for now it only runs on Windows."

*Submitted to passiveaggressivenotes.com*
posted by Brandon Blatcher at 10:35 AM on August 23, 2008 [4 favorites]

You know they were imagining the incompatibility note read out loud in their heads in the sneeriest John Hodgman voice they could muster up.
posted by Space Coyote at 10:46 AM on August 23, 2008

I don't have any machines that would run this.

We already saw this before, repeatedly. Nothing new here, I flagged it.
posted by lupus_yonderboy at 10:49 AM on August 23, 2008

Flagged as Single-OS-web-apps-are-soooo-1998
posted by DU at 10:57 AM on August 23, 2008

Works fine for me. Cool post, thanks!
posted by newfers at 10:58 AM on August 23, 2008

*Submitted to passiveaggressivenotes.com*

I wish I was cool enough to appreciate this joke.
posted by JHarris at 11:08 AM on August 23, 2008

it works fine on my computer and i love it. there's a strange dream like quality in the way it's collapsed a series of discrete episodes into some shifting, kaleidoscopic, temporal moiré. synth some shots taken over a couple of decades and it would be a fascinating way to travel back in time. i also can't wait for the porn industry to get their hands on this - multiply porn studio and extensive archive heat over photosynth et voila, a genealogy of desire.
posted by doobiedoo at 11:15 AM on August 23, 2008

No need to flag this one as this is a new development. Signal should have pointed out that this tech demo finally allows you to upload your own photosets, where previous versions only let you see already rendered sets.

What makes this different than other photostitching apps is how it attempts to reconstruct 3d space from a set of photos. One interesting thing you can do with detailed layouts is hold down the control key and see the 3d plot it created from the photographs. Here's some very good plots:

Grand Canyon
Eldorado Canyon
Boulder Dushanbe teahouse

The teahouse and Eldorado Canyon are both especially impressive when viewing the 3d plot. Maybe I'm easily impressed, but I loved that even the paintings and the trees were captured well.

And just out of vanity, here's a couple I made from the photos I had lying around:

Utah Valley
Hobble Creek area
posted by pandaharma at 11:18 AM on August 23, 2008

If it doesn't work for you, or you hate Microsoft, we all want to hear about, please post and let the rest of us know. Thank you.
posted by stbalbach at 11:26 AM on August 23, 2008 [13 favorites]

That's great and all, but when are we getting Seadragon? That looks like a way more interesting bit of Microsoft-devoured tech.
posted by Pope Guilty at 11:27 AM on August 23, 2008

I tried it by dumping a bunch of Safeco Field shots into it. It's not bad, really, even given me handing it shots with disparate focal lengths taken hundreds of feet apart.
posted by dw at 12:53 PM on August 23, 2008

FWIW I'm kind of sitting on a cool iPhone post at the moment. I'll be looking forwards to seeing similar complaints when I post it.
posted by Artw at 12:54 PM on August 23, 2008

And thanks for yours, pandaharma. When I was an undergrad at CU, the teahouse was still in crates waiting for someone in town to figure out what to do with them or where to put the teahouse. Now I wish they'd put it up while I was there -- that place is fkn gorgeous.
posted by dw at 12:55 PM on August 23, 2008

Wow. The teahouse one is really neat. Thanks pandaharma!
posted by Artw at 1:16 PM on August 23, 2008

[A few comments removed. Cool it.]
posted by cortex at 1:28 PM on August 23, 2008

This actually reminds me of the photo searching tool in Blade Runner, at least a little bit. It's cool, but it makes my brain hurt.
posted by selfnoise at 1:50 PM on August 23, 2008

Ok, I just went into my backyard and tried this out, and let me just say:

WHOA. WHOA WHOA WHOA.

I envision being arrested by homeland security in the near future. "Sir, why were you taking hundreds of pictures of that bridge?"
posted by selfnoise at 2:38 PM on August 23, 2008

I have this image of people walking through the mall, seeing a games store, and just shaking their fist in infuriated anger at all the games that don't work on their computer.
posted by smackfu at 3:02 PM on August 23, 2008 [3 favorites]

I am the director of Live Labs, the team that produced Photosynth (with a lot of help from MSR, UW, Virtual Earth, and others). I'd be happy to answer questions about any aspect of this project. The only caveat is that my responses will not be anything close to realtime because I have family visiting at the moment and I am also having to do some weekend work associated with the Photosynth release. But if you can tolerate a little delay, I'll do my best to answer.

On the whole "cool enough" comment, I can assure you that it was really meant to be more cute than passive aggressive. I really admire a lot of Apple products (as do many people from my team). In fact, I hired the designer behind OSX's dock to be the design director of the entire lab. I've also personally written over 100K lines of open source code. So while you may find it tempting to cast the world in something like black-and-white terms, I think the more you know about us the more you'll realize the truth is far more subtle, complex, and interesting.
posted by dr.flakenstein at 7:00 PM on August 23, 2008 [4 favorites]

Thanks for checking in, dr.flakenstein, and congratulations.
Questions: Where is the heavy lifting (ie.: photo matching) done, on the user's machine or on your servers? Will this always be web-based or will there ever be a standalone application?
posted by signal at 7:32 PM on August 23, 2008

I hired the designer behind OSX's dock to be the design director of the entire lab.

I'm picturing one of your team members in the lab kitchen. He reaches for the microwave handle to his right, but the microwave moves left and he misses, accidentally turning on the toaster instead. He grumbles, "Damnit, old man dr.flakenstein, why did you hire that asshole."
posted by scope the lobe at 7:33 PM on August 23, 2008

dr.flakenstein, don't obsess about the passive-aggressive thing. Here in the blue, we be swimming with the snarks. (And I don't see how it's passive-aggressive at all, really.)

In any event, I look forward to trying this out at Dragon*Con next week, if I can borrow a digital camera in time....
posted by JHarris at 7:38 PM on August 23, 2008

So is everyone dumping in giant images, or sizing them to something reasonable first? I was going to try one, but the photos added up to over 100MB.
posted by maxwelton at 7:46 PM on August 23, 2008

signal: Thanks for the kind words. The vast majority of the heavy lifting (in terms of the machine vision algorithms that actually derive the 3D structure) run on the client machine. A little known detail of the technology preview from two years back is that we needed weeks of CPU time to produce those first synths (but it was a one-time computation). A lot of the work done since that time was to speed up those algorithms so that now, typical synths of dozens of photos only take a few minutes on typical hardware, and synths with hundreds of photos are possible, but may take an hour or so of time. Nonetheless, for most people, the network upload takes longer and since we do both in parallel, it really is a win for everyone. Without pushing the computation to the edge, I am not sure that we could have released this for everyone so soon.

Since wrapping up the release, we "graduated" the project and it will become part of the MSN / VE family of products, so the future of it is out of my hands (in the sense that I can influence, but I will no longer own the major decisions). Nonetheless, I think this is a product that is vastly better as a hybrid, using client-side computation for the hardest stuff but everything else on the web. As a result, sharing (and the eventual combining of synths) is very easy and natural. In my mind, that's one of the most important design decisions that we made.

JHarris: no worries. I am well-versed with the blue. I addressed that one issue just because it's come up several times. Frankly, I am surprised that anyone actually reacted as it was just a throw-away joke for us.

maxwelton: Big images will obviously take longer to upload, but I would really encourage you to go as big as you can. The algorithms will work just fine on large images, and the experience of viewing it with our plugin (based on Seadragon) will be stunning. Some of the synths that we produced (e.g., National Archives) have document scans in the 50-70 Mpixel range (e.g., Declaration of Independence).
posted by dr.flakenstein at 9:15 PM on August 23, 2008

Thanks, Dr. I'll be curious to try it. I suspect the photoset I have isn't particularly synthy, though...with photosynth in mind, it obviously makes sense to take a lot of "transition" photos that may not have artistic merit but help to link nice photos to each other. If I'm thinking about this correctly.
posted by maxwelton at 9:32 PM on August 23, 2008

dr.flakenstein: I remember the TED demo used a collection of tagged images from Flickr to create the synth of Notre Dame. The public release uses only collections of user uploads. What kind of efforts are being made to link into large repositories (like Flickr, Picasa, or even internet-wide image scraping)? Since photosynth isn't just a method of presentation, but potentially a way to search for photos, do you think that photosynth will someday be able to integrate images without close regard to license? In other words, will it become a sort of web browser instead of a suped-up VR web app?
posted by cowbellemoo at 10:03 PM on August 23, 2008

I tried a collection of photos, and while some stitching happened, it's kinda interesting the ones that didn't--in a couple of cases, slightly different views of the same thing. Is there any room in future development to allow users to help the program? It can do its automatic thing, and then if there are stray photos or more than one 3-D group, perhaps the user could be prompted to click on a couple of shared points to help the program on its way?

Still pretty neat.
posted by maxwelton at 10:35 PM on August 23, 2008

Here's mine, of Trevi Fountain. I only wish I was thinking more of Synthing when I was in Italy. Sigh.
posted by disillusioned at 2:55 AM on August 24, 2008

disillusioned, the 3-d model of your set is pretty darn cool.
posted by maxwelton at 3:51 AM on August 24, 2008

To fully appreciate the neatness of this, check this set out (screenshots of Halo rather than actual photos):

http://photosynth.net/view.aspx?cid=820f7ee5-2657-483b-8f57-16ca5079eb87

Then hit P and explore the geometry of the level, recreated in 3D from 2D snapshots.
posted by selfnoise at 6:37 AM on August 24, 2008

maxwelton: Take a look at our guide which has some good tips on how to take photos with photosynth in mind. It does take a little bit of practice, but I think you'll like the rewards for your efforts.

At some point, I think we'll have to offer some basic editing capability so that users can remove photos from a synth or even adjust the aggresiveness of the matching algorithms so that they have the option o fremoving false positives. That said, we decided to keep the interface as simple as possible because of the newness of the whole concept.

disillusioned: I think you synth is quite good and I hope you try for more.

cowbellemoo: Besides the user created synths, we also launched with partner synths (like National Geographic). However, there is little right now in terms of crawled content. In truth, we had wanted to launch with the sort of collections that you mentioned but that work was lower priority and woudl have delayed our launch, so we went with what we had. I suspect that you'll start to see those sort of synths appear within months (if not even weeks).

Finally, and apropos noting at all, here's a synth of our "war room" on launch night were we were monitoring everything world-wide. It took 40 minutes for the first porn synth to appear. And our database for metadata started to burn up within 70 minutes post launch. But still, it was a great night for us.
posted by dr.flakenstein at 8:32 AM on August 24, 2008

Is there any room in future development to allow users to help the program?

That's my biggest problem. It's just a black box, and if the results don't come out how you like, that's life. I have panoramas where it only managed to stitch together two sets of three photos, with no reason why it couldn't do the full thing. And I have some others where it's clearly just placed a photo wrong, and there's a huge jump when it changes to that one.

I think what wasn't obvious to me with the earlier posts was that "stitching 150 random photos" wasn't so much a feature as a necessity. If you only have 15 random vacation snaps, it just doesn't work. It can do panoramas, sometimes, and it can figure out when you zoomed in, but that's about it. You need your 15 snaps and 10 other people's.
posted by smackfu at 9:08 AM on August 24, 2008

A few examples:

A complete failure at doing an object

A 100% synth with random vacation photos where a bunch are just misplaced

For some reason, Chichen Itza synths always work
posted by smackfu at 9:11 AM on August 24, 2008

I just shot a 3D view of a local monument, which took about 110 photos in total. (DSLR buffer helps a LOT here). Once you get used to being somewhat cautious about coverage, the program has a pretty good success rate. I would definitely like to see the option to just remove the shots that don't get stitched, at least.

Also I can't for the life of me figure out how to geotag synths... there's a button to do it but it just takes you to Live maps and then... what? There's no "object is here" button to finish the process out.
posted by selfnoise at 9:14 AM on August 24, 2008

The cup lacks texture, which is probably part of the problem. The lady in the instruction video warns against white walls. Also (although this probably isn't much of a factor in that object attempt) I would think you would want to avoid any shallow focus.

Here's my successful 3D object (successful at creating one, anyway, the exposure was pretty inconsistent.)
posted by selfnoise at 9:18 AM on August 24, 2008

It's funny that your monument has a fantastic point cloud, but switching between photos is still really wonky. Especially if you click around the torus to change the angle, stuff just fades away rather than properly transforming like you would think it would be able to.
posted by smackfu at 9:28 AM on August 24, 2008

Yeah, the transition from cloud to 2D framed images is jarring on the 3D surround objects. I think the holy grail would be the computer somehow using the photos to divine the geometry, then cut the photos up and texture them onto the geometry, ala a FPS engine.
posted by selfnoise at 9:36 AM on August 24, 2008

I suspect that you'll start to see those sort of synths appear within months (if not even weeks).

Yay! I can't wait!
posted by cowbellemoo at 10:41 AM on August 24, 2008

The random errors are pretty tiresome. Does the server actually do anything in this process? Or is it just to share with others? It seems like a huge weak link. I've had only about 50% success rate in synths, between the Error A00000B and the hanging at the last step. And of course it only fails 20 minutes into the process. Why can't this just be a program?
posted by smackfu at 10:47 AM on August 24, 2008

Finally decided to install this and yeah, it's neat. I thought from the advertising copy it was yet another VR panorama photostitch thing. But it's not, it's a much more interesting sort of 3d photo album. I particularly like that it highlights the original photos, isn't focussed entirely on synthesizing some 3d view from images. The navigation transitions are nice, it makes me feel like I'm looking through an old View-Master.

All the navigation controls on the UI make me think I'm expected to interactively fly through the scene. I find it works better if I just sit back and hit space to watch it, like a slideshow.
posted by Nelson at 10:47 AM on August 24, 2008

Also: paging David Hockney.
posted by Nelson at 10:49 AM on August 24, 2008

I still liked the passive-aggressive joke.
posted by Blackadder at 12:45 PM on August 24, 2008

I still think that letting the user identify three points in common (or more) on any two images would allow the software to triangulate the image into an existing 3-D space, wouldn't it? That would make it far handier with vaguely connected but not intensively so sets of photos.

I threw this one up. It did pick out a few 3-D groups. But it you look in the un-attached images, you can identify several that are all slightly different views of the same subject...take the yellow airplane, for example. There are a few shots of that when viewed from a human eye, would seem to be natural matches for something like this.

Anyway, be interesting to watch developments.
posted by maxwelton at 3:56 PM on August 24, 2008

It's really fascinating to see what people have uploaded, and what that says about user hopes or expectations. From a single snapshot ("ma, you're gonna be 3-D!") to stuff like the set I put up, which I recognize as not being anywhere near detailed enough but is probably more indicative of what a typical set of "touristy" photos looks like.
posted by maxwelton at 4:09 PM on August 24, 2008

smackfu: I am actually a bit confused now. Your synth A 100% synth with random vacation photos where a bunch are just misplaced is actually really good. I didn't see a single photo misplaced. It is true that some of the features within photos are not consistently posioned relative to one another, but that's because there is no way of making them line up (you are, after all, rotating around some key features and the angles are such that positions between tripples of things actually changes). But even in these cases, I can't see a single photo that I would call "misplaced".

You synth of the coffee mug didn't work because there were not enough features for the machine vision code to identify across photos. When it did work, it was on the on the mug itself. But the actual mug and table top is simply too uniform for it to work. There's a chance that something could have been recovered with more photos that had more overlap (in terms of both space and angle), but you never know until you try.
posted by dr.flakenstein at 6:00 PM on August 24, 2008

smackfu: I am sorry that your synther is hanging at the last step. The A00000B error usually happens when your upload get interrupted because of lost connectivity. The long hang during the publishing phase seems to happen after on the next reattempt. We are actively trying to track down where the block happens right now, but (and I know this will be an unfullfilling answer) for now the third time seems to be the charm.

Re. your question about the server: yes the server is mostly for hosting the synths but this involves more than you may think. There is the point cloud, the image tiles (which allow for the experience to be smooth atvarying scales) and the metadata that pull it all together.

Yes, it is possible to make a stand-alone application but, in truth, I don't think we could have made it a free product. We are pushing the data into a central location so that we can eventually stitch synths together. If the product was designed so that everyone was encouraged to make little "island" synths, then there would be little hope that all of the different synths could be combined into a larger, unified, collection. Our sincere hope is to make a system that continuously improves because of user contributions, thereby increasing the value for everyone and in a highly non-linear way (think about how your vacation photos and my vacatio photos are, well, just okay, but together they are vastly better. Now multiply this by million of people.)
posted by dr.flakenstein at 6:08 PM on August 24, 2008

I did a gazebo today waiting for laundry.

The inside/outside trickiness of the pictures was handled amazingly well by the software. The navigation is really tricky, though... I really want to just move around with the arrow keys in the 3D cloud but once I get inside the gazebo I just get stuck there.
posted by selfnoise at 7:26 PM on August 24, 2008

But even in these cases, I can't see a single photo that I would call "misplaced".

Thanks for the responses, Doctor. I guess I was calling it misplaced when it really is placed in the right place, which is impressive, but oriented in 3d space wrongly. It's jarring when you switch to a photo and it seems to rotate in the wrong direction while doing a crossfade.

This one of my desk came out quite well once I finally got it to upload. First I did 30 photos and ran it, then I took another 30, then I just went nuts and took another 60 for a total of 120. It's surprisingly hard to take that many photos, since we're used to taking photos that are meaningfully different, and photosynth really needs photos that are meaningfully the same -- except for a little bit. The resulting point cloud is quite nice.

It's really a fun little toy. Next time I'm on vacation, I might just saturation bomb some locations with photos.
posted by smackfu at 7:47 PM on August 24, 2008

You synth of the coffee mug didn't work because there were not enough features for the machine vision code to identify across photos.

Interesting. I guess I expected it to work by looking at the edges to find points to correlate, so in this case, my coffee cup had strong edges, and the table had strong edges and corners.
posted by smackfu at 8:06 PM on August 24, 2008

Selfnoise, it's interesting that some shots ended up sideways. If you click "next 3-d group" you can escape from inside the gazebo.
posted by maxwelton at 8:07 PM on August 24, 2008

I'm getting interesting results setting my camera to 'continuous shot' and a low resolution.
My cat sitting on a practice amp.
posted by signal at 11:56 AM on August 25, 2008