3DeeSlide & 3DeeCentral review

[Fredz]

I just received my 3DeeSlide for free thanks to Spatial View limited time offer, so it's time for a review of both the 3DeeSlide device and the accompanying 3DeeCentral application. Be prepared for a long read...

3DeeSlide package.jpg

At this time it's only compatible with the iPhone 3GS and the iPod touch 3rd generation, support for the iPhone 4 and iPod touch 4th generation will be announced at a later date as specified on Spatial View website. The 3DeeSlide normally costs $9.99 and the 3DeeCentral application is free to download on the App Store.

I tested the application and the device on an iPhone 3GS.

3DeeSlide

The package

The device was packed in a 5"x7" bubble envelope, it didn't suffer from travel and could easly fit in my mailbox.

The package included :
- a 3DeeSlide ;
- an envelope containing one 3DeeLens ;
- a user guide.

The user guide is very short (one little sheet of paper) and does only explain how to mount the device. Simple and to the point, but nothing more.

The 3DeeSlide is made of hard black plastic and looks like a protecting case, you'll first have to slide your iPhone or iPod touch into it. It features openings so you can still switch the volume on/off or change the level, you can also plug your device with the USB cable to charge it or connect it to your computer.

The 3DeeLens is a small flexible lenticular sheet that you'll slide into the 3DeeSlide once it is attached to your device, with the bumpy side facing up.

First impressions

Overall the 3DeeSlide and 3DeeLens look a bit cheap but at $9.99 it's no surprise, and since I got them for free I would be hard-pressed to complain.

The touch screen is still very responsive when the 3DeeLens is installed which is a very good point. It seems that the problem with the previous version of the device (3DeeShell) which pushed Gizmondo to give it a bad review is no longer relevant with this update.

Although I've read that it should work in both horizontal and vertical orientations, I could only test it in the horizontal one. I didn't find any content available for the vertical mode and the screen doesn't rotate in the 3DeeCentral app when you rotate the phone. Perhaps a new version of the 3DeeCentral application will address this in the future.

Adjusting the screen

The lenticular screen is quite easy to position but the procedure is quite cheap, when you move it horizontally it has a tendancy to also move vertically. It's also a bit difficult to obtain a precise positioning since it's moving jerkily, so it takes some time to obtain the best possible alignment.

Forget about trying to obtain a perfect alignement though, I think the best solution is to have one that's good enough and then simply move the phone relative to the eyes to obtain the best effect.

You also need to have the sheet perfectly affixed on the surface of the screen or you'll see some ghosting or fuzziness in zones where the space between the sheet and the glass is bigger.

Cleaning

It's easy to add finger traces on the sheet so you can be tempted to use some cloth dampened with water to remove them. But don't make this mistake or else you won't be able to dry the sheet easily afterward. It will also be harder to place the 3DeeLens on the screen because of the adherence caused by humidity. And if you still manage to put it on the screen you'll see traces of humidty where there is perfect contact, although it's a lot less visible when viewing 3D content than with the screen switched off.

3DeeSlide traces of humidity.jpg

Photo taken after many attempts at drying the sheet and replacing it on the screen.

3DeeCentral

The purpose of the application is to view images and videos which can be downloaded in the store section, free and paid content is available. I only tested free content for now since there is already a decent selection available (2 sets of images and 14 videos).

I had previously installed 3DeeCentral and already downloaded some free images and videos, but the day I received the 3DeeSlide I discovered that I had lost all my downloads when launching the app again (I also had to retype my password). Fortunately they were all available in the download list so it was just a matter of waiting for them to be downloaded automatically, but it's still annoying to have to wait 10 minutes before being able to use a device you just received. Quite a bad start indeed.

The interface

The interface is quite simple and easy to use but it's only in 2D so the texts are almost unreadable when the 3DeeSlide is mounted. Removing the device to search content and download it to only put it again for viewing content is really a tedious task and a complete waste of time. That's quite a gross overlooking on Spatial View side and I can't understand why they didn't also provide a 3D interface.

The commands are quite slow and unresponsive in the main page of the store, when you tap on a button you often have to wait some seconds before anything happens. Sometimes you can see a login screen during these periods, so I guess the application may have to contact a server to authenticate the user for each action. It renders the experience quite frustrating and I really see no reason why they decided to do this.

The application is also a bit buggy. It crashed on me several times while simply launching the app or tapping on a button, but it didn't happen very often. It also froze one time while playing a video but didn't happen again after prolonged use.

Several times I've found non-free content in the free section of the store, that's pretty dangerous so you'd better have a second look to the price displayed before starting a download. I really hope that it hasn't been done on purpose by Spatial View to sell more content, that would be quite a mean thing to do.

Overall the application feels like it's in a beta stage, it should have been tested more thoroughly before release.

Calibration image

3DeeSlide calibration image.jpg

Each time you want to view content, a calibration image is first presented on the screen to help you adjust the viewing angle, which is quite a nice idea. You can then move the device to minimize ghosting and fuziness before you start playing content, but a very small movement can ruin the effect in some parts of the image or all of it.

I thought autostereoscopic screens were a lot less sensible to the viewing position and I've often heard that you had some margin for the "sweet spot". Unfortunately I've verified that it's not the case, maybe it's due to the small size of the screen. Note that looking at real content is more forgiving though, a not so precise alignement seem to be less disturbing than when you look at the calibration image.

I could never obtain a crisp view of the overlay interface (cross and play button) while looking at the calibration image. Although it seems it has been adapted for 3D, there still must be some misalignement in it. One thing I don't understand is why they felt the need to put an accompanying text in 2D, it's really not informative at all and it's quite difficult to read this way.

Viewing content

Each time you want to view an image or a video, the navigation interface is displayed and stays on the screen for two seconds. It's not a very long delay but it's still quite annoying because it disturbs the depth perception of the stereoscopic content that is presented.

When you are viewing an image you can tap on the screen to hide the navigation interface, tapping on it again will make it reappear. You can view images in diaporama mode by tapping on the play button, the interface will then disappear and the images will be displayed in intervals of approx. 3 seconds.

You can also tap on the left or right arrow to go to the previous/next image, but in this case the interface is displayed on top of the content and distracts the depth perception. You can also slide your finger left or right to navigate between the images to not show the interface although it isn't mentionned in the user manual nor the interface.

Free images

The selection of free images is average, the Slam Comics serie is pointless (2D drawings converted to 3D, giving a sort of cardboard effect) but in the 3D Images-Sample selection there were some good shots. Some of them didn't have much visible depth and one of them was even inversed, so you'll need to move the device to view the serie correctly. Some images also exhibited some window violations which are particulary disturbing on such a small screen. That's really a bad way to demonstrate this device.

The resolution does suffer a little bit but not as much as I expected considering the technology that's been used. On images taken quite close (flower, bull) it's not very noticeable, but it's a little bit worse on images taken from a distance (garden, field, city). I could also barely see the diagonal pattern when the alignement was correct, which is really a good point.

[youtube]http://www.youtube.com/watch?v=YjrSIOKxb3w[/youtube]

Overall these images can give a good idea of the capabilities of the device, but the content should have been chosen and edited more carefully. In this state, I'm not sure it will really convince people of buying new content.

Free videos

There is quite a good amount of free content to view with the 14 videos available (only 13 in fact, one of them is proposed in both English and French versions). With some videos I found very disturbing the fact that the first few seconds only showed a black screen, it was easy to loose the perfect spot during this time before the real content was displayed.

In several videos the frequent changes in depth and camera angles between shots were quite unpleasant (particulary visible in Human Flight 3D Trailer), there were also quite a lot of frame violations and sometimes it looked like the whole image had been pushed a little bit out of the screen to give an overall pop-out effect. That was not the wisest choice to make I'd say.

I particulary liked The Ultimate Wave Tahiti trailer, the aerial scenes in Human Flight 3D Trailer and live action scenes with CGI elements in the trailer of Legends of Flight. I couldn't read the ANA pilot scene video, it stopped each time with a message saying it could not decrypt the file. There was no option to download it again or to delete it, so I guess I won't be able to view it as long as the bug is not corrected in the 3DeeCentral application. Also the Shine Music Video was inverted.

I'd say the best content was the one provided by professionals, in comparison the other videos were somewhat lacking in several departements (depth budget, interest, vertical parallax). There were some pop-out moments that looked good, but the overall depth was a lot less impressive than with my shutter glasses and my 120Hz display.

I'd like to compare this autostereoscopic screen to my shutter glasses installation using my collection of 3D trailers, but I'd need them to be available in the 3DeeCentral store or I should find a way to encode them for this device.

Health considerations

Viewing content when you are not in the perfect spot is a little bit disturbing and fatiguing, no matter what you try I'm afraid it's going to be quite difficult to avoid that.

After the first use my eyes felt a little bit tired, for several minutes I also had a weird feeling while looking at objects in my environment, probably related to focusing I'd say. Maybe it was because I didn't sleep much the previous night, so you should maybe not try using the device if you feel a little bit tired.

The trouble was less evident after several uses though, I guess my eyes or my brain got used to it after some time. I guess the fact that a correct alignement is very hard to obtain is certainly the main factor, being presented not perfect 3D must surely be the cause of such a fatigue.

Conclusion

Overall I found the experience quite enjoyable even if I've mixed feelings about it. I was really disappointed by the very limited viewing angles which I really didn't expect with this kind of technology, but I was favorably surprised by the not so apparent loss of resolution (depending on the content) and the diagonal pattern that is barely visible. The ghosting seemed to be quite low too - really better than what I did expect - but I'd need to create a ghosting test (based on the DDD one for example) before being able to confirm this.

Still it's leagues behind stereoscopic technologies like shutter glasses, and I don't see this type of glass-free technology replacing the ones with glasses anytime soon considering all the pitfalls this solution has. Once the problems with the application and the content are resolved, I guess the experience will be a little bit better, but problems inherent to this technology will nonetheless still be present.

I'd also like to test the GRilli3D or the Nintendo 3DS which use parallax barriers to compare the different glass-free options, but for now I can't really say that autostereoscopic technologies are ready for prime time. I will continue to test the device though and I will also probably try to adapt/create some content or write some interactive 3D applications that make use of it.

Anyway, since this was a free ride I can hardly complain, and even at $9.99 I'd say it's a nice technology to experiment.

[cybereality]

Sounds interesting. I wasn't expecting wonders, but maybe I hoped for a little bit more. I am pretty sure that commercial level auto-stereo is better than this, like the Nintendo 3DS or Fujifilm's W3. However I have not seen any of this in person, just my DIY parallax barrier, so its hard to be certain. It seems like a lot of the complaints you have stem from them having poor or badly authored 3d content. Its possible with better content it may look nice. And, for $9.99, what do you expect? Also, can you explain more about the process of taking the lens off and on. Is this easy to do? Does it snap into place, or do you have to adjust it every time? I am thinking about building something like that for my DIY parallax barrier, so its easy to take on an off. Still have some stuff to figure out, though.

[Chiefwinston]

Thanks fredz, I'll give it a try on an iPad once they release the screens. For now I can just sit on the sidelines and wait. Thank you for your time and review.
Cheers everyone

[Fredz]

Sounds interesting. I wasn't expecting wonders, but maybe I hoped for a little bit more. I am pretty sure that commercial level auto-stereo is better than this, like the Nintendo 3DS or Fujifilm's W3. However I have not seen any of this in person, just my DIY parallax barrier, so its hard to be certain.

Yes, the 3DeeSlide is quite cheap and I also expect better from the Nintendo 3DS or Fujifilm cameras, but I think they can only give a more precise alignement, the limited view angles should be similar I guess.

It seems like a lot of the complaints you have stem from them having poor or badly authored 3d content. Its possible with better content it may look nice. And, for $9.99, what do you expect?

Yep, I didn't expect that much for $9.99 as I said, but it's a pity they didn't prepare the launch of their device and software a little bit better. Having such basic errors such as frame violations and inversions shouldn't have happened. We'll see if they'll correct this in the future and add more interesting content, like 3D trailers/clips of blockbuster 3D movies.

Also, can you explain more about the process of taking the lens off and on. Is this easy to do? Does it snap into place, or do you have to adjust it every time? I am thinking about building something like that for my DIY parallax barrier, so its easy to take on an off. Still have some stuff to figure out, though.

Take a look at this video, the guy shows how to put the lenticular screen at 0:50" :

[youtube]http://www.youtube.com/watch?v=XUcUlxpkIzQ[/youtube]
You can do it quickly as he does but it can take some time to adjust it correctly so the calibration image looks good. Unfortunately it's quite difficult to show this on video, I would need a stereo camera to give a good illustration of the operation.

[Fredz]

It seems the ANA video has been corrected, now I can read it without a problem. It's a clip from the mexican feature film of the same name and it's a cute little animation, very well done. I've also downloaded all the free movies now and among them the Slam Fighter 3D short makes a good use of 3D, one of the best examples for now.

The other videos are not so good, some have vertical parallax (Twin Dragons, Shine music video), others seem to be 3D conversions (3D Sun Trailer) and many have mild 3D effects (Safety Geeks, Elephants Dream, Sunflowers, Las Vegas Nights) or are of a doubtful interest (Reminiscence).

From the 14 video available, I found these ones quite good : Slam Fighter 3D, Human Flight 3D, The Ultimate Wave Tahiti, Legends of Flight and ANA.

[Fredz]

Also, can you explain more about the process of taking the lens off and on. Is this easy to do? Does it snap into place, or do you have to adjust it every time?

Now that my 3DeeLens is dry, I can confirm that taking it off and on is quite a fast operation. You still have to adjust it a bit to minimize ghosting but with some practice it can be done quite fast too.

However I still prefer to navigate in videos and images lists with it on because it's still quite annoying to have to replace it each time, even if it makes the interface harder to read. Something to take into account if you want to write an interface for your DIY parallax barrier.

[Fredz]

Ok, I've done a little bit of research and I think I've identified the approximate specs for the lenticular screen. I've then implemented a little app to do some tests, it's able to convert stereo pairs into the interlaced format needed by this device.

Here is an example of what I obtain for the DDD ghosting test :

3DeeSlide - Left.jpg

3DeeSlide - Right.jpg

[youtube]http://www.youtube.com/watch?v=EKB19R3Hu6w[/youtube]
Note that the photos doesn't really do justice to what it looks like in reality, it's quite difficult to hold a camera still while pointing it at such a smal device. And my camera doesn't have a very good macro mode so it does look a little bit blurry too.

My actual implementation of the interlacing is not very advanced either, I'm not completely sure of the tilt angle I calculated and I didn't take subpixels into account yet (nor the black space between them). I guess with some more work I should be able to enhance it and reduce the ghosting, banding and color artifacts. I should also be able to render more than two different views but that'll need some more work too.

Here is a video showing a red and blue image, the banding is quite awful in this one, but I think it does still give an acceptable result :

[youtube]http://www.youtube.com/watch?v=noqmJfNaS74[/youtube]
I'll continue to work on this and keep you informed of any progress, even if it seems that this device does not interest a lot of people...

[cybereality]

Hmm, that looks pretty good to me. I mean, those DDD ghosting shots look reasonable. I wouldn't expect it to have no ghosting. With the right content, it might actually be pretty nice. I am planning on doing some experiments with my DIY parallax barrier using Android phones/tablets. So I would be happy if I could reach that level of quality.

In terms of commercial auto-stereo displays, I am hoping they are better than these ghetto solutions. Well, I know the viewing angles are small (like on the Nintendo 3DS) but the image quality should be much better. Also, the Fujifilm W3 doesn't use a parallax barrier, it uses a directed light system. So it has the potential for much wider viewing angles than are possible with parallax barrier (although the current implementation might be similar).

[Fredz]

Hmm, that looks pretty good to me. I mean, those DDD ghosting shots look reasonable. I wouldn't expect it to have no ghosting. With the right content, it might actually be pretty nice. I am planning on doing some experiments with my DIY parallax barrier using Android phones/tablets. So I would be happy if I could reach that level of quality.

I'm quite confident that the quality can be enhanced in a significant manner with some antialiasing and subpixel rendering, some more tests to find the real angle of the sheet should help too. I may also try to support multiples views if the resolution doesn't suffer too much, at this time it's quite annoying to have to stay in the sweet spot. Another thing I'd like to test is the accelerometer to adapt the view in realtime according to the phone's movements.

In terms of commercial auto-stereo displays, I am hoping they are better than these ghetto solutions. Well, I know the viewing angles are small (like on the Nintendo 3DS) but the image quality should be much better.

It depends on the resolution of the underneath display anyway, even with such a cheap solution like the 3DeeSlide I think it would look really great on an iPhone 4 (4x the resolution), it's still the biggest resolution display for this size (3.5"). And with the integrated face camera it may be possible to have a tracking solution in real-time.

The other solutions that are going to arrive should still be better in term of integration, like you won't have to remove the barrier and it'll be perfectly aligned, but in the end that's the resolution that matters.

Also, the Fujifilm W3 doesn't use a parallax barrier, it uses a directed light system. So it has the potential for much wider viewing angles than are possible with parallax barrier (although the current implementation might be similar).

Do you know what's the exact system that the W3 uses ? It looked a lot like lenticular screen or parallax barrier from screen caps I saw, you can even see a vertical moiré in one of them :
http://www.flickr.com/photos/firrs/5185 ... otostream/" onclick="window.open(this.href);return false;
And from what I read the viewing angle was very narrow too.

[cybereality]

Do you know what's the exact system that the W3 uses ? It looked a lot like lenticular screen or parallax barrier from screen caps I saw, you can even see a vertical moiré in one of them

Yes, you are right. It does use a lenticular screen: http://www.fujifilm.com/products/3d/cam ... fications/" onclick="window.open(this.href);return false; (scroll down to 'LCD Monitor').

However the older version, the W1, did employ a "Light Direction Control" system: http://www.fujifilm.com/products/3d/cam ... index.html" onclick="window.open(this.href);return false;

Not sure why they would go to lenticular on the newer model, maybe this light direction control isn't that good. I can't seem to find any information on it, however I did read a explanation of it back when the W1 first came out. Its almost like a shutter glass solution, but without the glasses. Basically the screen has 2 backlights, one angled to your left eye, and one that is angled right. There might be an additional layer behind the LCD panel that helps collimate the light. Then the screen does normal page-flipping (I guess at 60Hz) and each frame the backlight is synced to light only the correct eye. So I believe it is a full-resolution solution. Not sure why they abandoned it (maybe it was too expensive). I thought it had a lot of promise.

[Likay]

I have the impression they still use it in the w3 (eventhough i haven't seen the w1). The display flickers in 3d mode while it's experienced static in 2d-mode. The tech with directed backlight is actually a tech based on lenticular technology as well.

[Fredz]

I've recalculated the angle and I think the images are looking a little bit better, but it's still not perfect, I need to do some more tests to find the best value. Currently the value I'm using is 33.19 and should correspond to the value used by the 3DeeCentral app. But I think it's still not perfect, I'll try to create a pitch test to estimate the perfect value. Previously I was using an angle of 33.5684 which produced a different and better looking moiré than with the value I use now, but I needed to repostion the lens each time I was using the 3DCentralApp so I've settled on a more compatible value.

I've also implemented a multiview conversion, it works more or less but the result is not as good as I expected. I only tested it with 5 images using uniform colors (red, yellow, green, cyan, blue like in rainbows) and I can see the color changes when I move the phone or my head, but I rarely obtain a pure color, it's mostly a mix of 2 or 3 colors but with a dominant one. I suppose the problem comes from the not perfect angle calculation, but it may also be improved with subpixel rendering. Before implementing that I'll try with a 3D model rendered in Blender for 5 views to see if the effect is good with real content.

I've also implemented video conversion using Avisynth. It was the first time I used this application but it took me less than an hour to write the script, and in the end it's only 6 lines long. This software is really wonderful, what a shame I didn't try it before.

I've converted the 3D trailer of Disney's Tangled, the depth is quite good and there are some great pop-out moments, like when the arrows hit the tree. But the resolution is clearly lacking, I hope I can correct this with subpixel rendering or by modifying the parameters I used to convert the movie with MEncoder.

[youtube]http://www.youtube.com/watch?v=lYa0J6OA6eY[/youtube]

Note that the sound is completely out of sync, I must I've done something wrong. And the quality is not as good as with the original MP4 I encoded, blame YouTube on that... But you can still watch it directly on your iPhone/iPod and experiment the stereo 3D effect.

If you're interested here is the script I wrote (for SBS-full movies, should be easy to adapt to other formats). I guess it can also be used as is for slanted parallax barriers :

interlacer.avs :

Code: Select all

video = DirectShowSource("path\filename.avi").LanczosResize(960, 320).ConvertToRGB32()

left = Crop(video, 0, 0, 480, 320)
right = Crop(video, 480, 0, 480, 320)

mask = Imagesource("blackwhite.png").ConvertToRGB32()
final = right.ConvertToRGB32().Mask(mask)
Layer(left, final)

You'll need this mask image to use the script, it does use the 33.19 angle value :

blackwhite.png

And here is the MEncoder command line to convert the video for the iPhone :

Code: Select all

mencoder interlacer.avs -o output.mp4 -lavfopts format=mp4 -faacopts mpeg=4:object=2:raw:br=128 -oac faac -ovc x264 -sws 9 -x264encopts nocabac:level_idc=30:bframes=0:global_header:threads=auto:subq=5:frameref=6:partitions=all:trellis=1:chroma_me:me=umh:bitrate=750 -of lavf

[cybereality]

Oh, now I want to see this myself. There is a iPod touch at work I can probably borrow, its the 3rd gen 8GB model which I guess should work. Would like to know how this looks.

[Fredz]

Yes the iPod touch 3rd gen is supported by the 3DeeSlide. You'll need to install the 3DeeCentral app on it (maybe with iTunes, don't know if they support WiFi) and register an account to get the device for free. Or you can buy it directly from Spatial View, afterall it's only $9.99...

[cybereality]

Man, I thought it was 3rd gen, its actually 2nd gen I have. Looks like it may not work. However I believe the display panel in the 2nd and 3rd gen models are exactly the same. So the 3DeeSlide thingy should still work. Maybe just their software doesn't work, which I can probably deal without. I was going to buy the thing for $9.99, but they want $20 just for shipping & handling. To hell with that. I'm gonna pass just on principle.

[Fredz]

Too bad, you can still try to install the 3DeeCentral on your iPod touch to see if you're eligible for the free 3DeeSlide. But I'm afraid these devices don't have enough power to interlace the images in realtime, I read that the guy who worked on the video player used OpenGL ES to do that.

By doing some research I've determined that they use a 42 LPI lenticular screen (42.02 according to my imprecise calculation), but I don't know the values for thickness or viewing angle. And I don't know if different values would have any influence in the end, apart for the viewing distance. A 60 LPI lenticular screen could give better results in terms of resolution, but I'm not sure why they decided to go the 42 LPI route.

[Fredz]

I've implemented subpixel rendering and also found a way to remove the moiré I got when using images in plain color (by supersampling). Here are the new ghosting tests which look a little bit better than before (I really need a tripod to take photos, lots of blur in those) :

DDD ghosting SS Left - 640.jpg

DDD ghosting SS Right - 640.jpg

Thanks to the supersampling a simple black and white image can now be used to easily calibrate the lenticular screen. The image with 5 views I tested before is also much better looking thanks to this addition, now I can see one plain color for each view. Next step will be the creation of a 3D model in Blender and render 5 views from it.

I've also created a new supersampled mask for the AVISynth script, but I need to recreate it programmaticaly because for now I've done it manually and I created moiré lines unintentionally.

Overall I'd say good progress, but I'm still not satisfied by my rendering. The perceived resolution is still much lower than with the content available in the 3DeeCentral app, particulary on lines that are almost parallel to the lenticular lines (upper right of the big circle in the ghosting test for example). I still don't understand what they've done to correct this, but I'm still searching...

[cybereality]

Yeah, that does look better.

[jgh]

Hey... this is a really interesting topic. I'm actually the guy that wrote the video player for 3DeeCentral... It's written using OpenGLES 1.1, although an ES 2.0 approach would definitely be preferable...but if I had done that it would have made your reverse engineering efforts really easy

Yeah some of the content isn't exactly "up to par" - but believe me the guys making the deals at SVI are working their butts off to try and bring in top-tier content. I think it's a bit of a challenge with the way the industry is with 3D (read: nervous about getting into bed with small companies)

The restriction on the older devices is because of power - yeah - there's some necessary (for 3DeeCentral) stuff going on in the background that prevents it from running properly on the iPhone 3G and 2nd gen ipod touch. If you were to write a video player using OGLES1.1 and FFMpeg (or AVFoundation thanks to 4.1) and get it to work, though, it would work. I promise you that. Those devices have enough power when you devote the power to just the video playback - and in fact the wall you hit is with texture upload before you run into problems with decode times. The post processing is negligible compared to colorspace conversion - which you can offload to a shader on newer devices, or if you get really creative you can do it with texture blending with ES1.1 - and decode.

It looks really, really nice on the new retina displays though - very clean and crisp.

Fredz, I wish I could help you but I fear I'd be in trouble We did a lot of work to try and get the ghosting down as low as possible, and it was quite a balancing act between touchscreen capabilities and image quality.

[Fredz]

Hi James, I found your resume some weeks ago on the Web and was tempted to contact you about interlacing, but I hesitated because of the possible conflict of interests. Seems I was right after all...

Anyway, I have lots of questions to ask about your work for which you shouldn't have to divulgate sensible informations, here they are if you don't mind answering them.

First I don't understand the reason why you did implement interlacing as a real-time process with OpenGL ES instead of using an off-line encoding for the videos. This way you could have made the 3DeeSlide and the 3DeeCentral application available for a lot more users (iPhone 3G, iPod 2nd gen), which I think is what Spatial View really needs to have the most viable business. Why did you make this decision ?

Another thing I remarked was that you didn't use anti-aliasing for the interlacing mask, this generates a diagonal moiré effect that is quite unfortunate, as can be seen on videos with an uniform background. It's especially visible in the Tahiti video when a white backgroud is displayed as shown in the following screen capture. Is there any reason why you did use a discrete mask instead of an anti-aliased one ?

moire.jpg

I guess you did use a mask similar to this one :

blackwhite.png

instead of one looking like this one, which is not perfect because I created it manually by copy&paste, but does not produce diagonal moiré :

blackwhite480aa.png

Another question I had is why you didn't create any game for the 3DeeSlide as was done for the 3DeeShell. I thought you were using an off-line interlacing that was not possible to render in a game in real-time, but since you've confirmed this to not be the case, I wonder why you didn't adapt the previous game (a space shooter IIRC) for the new lenticular screen.

I've studied the 3DeeSlide lenticular lens quite in deep and I don't understand why you didn't implement multi-view either, since it's slanted (33.69° or atan(2/3) more precisely) and it covers approx. 4.62 horizontal pixels. My first tests seem to indicate that it should be possible to do it, why didn't you take advantage of this, is it because the resolution in 3D would be too low ?

I'm still trying to obtain the best possible image quality and I'm currently stuck in the anti-aliasing techniques you seem to have implemented correctly on your side. I've read a lot of papers about this subject (Konrad, Agniel, Ruijters, etc.) but they're quite complicated and I lack the mathematical background to implement what they describe. Did you use the same techniques as the ones presented in these papers or did you use another solution ?

Thank you, and btw great job on your post-processing implementation.

[jgh]

Hi James, I found your resume some weeks ago on the Web and was tempted to contact you about interlacing, but I hesitated because of the possible conflict of interests. Seems I was right after all...

Anyway, I have lots of questions to ask about your work for which you shouldn't have to divulgate sensible informations, here they are if you don't mind answering them.

First I don't understand the reason why you did implement interlacing as a real-time process with OpenGL ES instead of using an off-line encoding for the videos. This way you could have made the 3DeeSlide and the 3DeeCentral application available for a lot more users (iPhone 3G, iPod 2nd gen), which I think is what Spatial View really needs to have the most viable business. Why did you make this decision ?

Real-time interlacing is feasible for the older hardware as I mentioned before, but because of some copy protection stuff it gets just a little too slow (i.e. there is some choppiness in playback) for the iPod Touch 2nd Gen. It's kind of unfortunate that we couldn't go with this generation, but I guess the decision was that eventually everyone would upgrade anyway. As such they only want to support 3rd gen and higher. I'm actually no longer with SVI (as I relocated to the west coast) so I don't have any insider info really, I imagine the iPod Touch 4 and iPhone 4 cases are on their way though.

Pre-interlaced content is what we did for some of the 3DeeShell content before we properly figured out real-time interlacing on the iDevices. There is no mask used as you are suggesting, and lenticular interlacing is more complex for LCD screens than it is for print. The down-side to pre-interlaced content is because of the interlacing, it doesn't compress very well. So even short video ends up becoming quite huge. So it's a trade-off. Do we abandon lower-end devices, or do we force their owners to download massive pieces of content?

Another thing I remarked was that you didn't use anti-aliasing for the interlacing mask, this generates a diagonal moiré effect that is quite unfortunate, as can be seen on videos with an uniform background. It's especially visible in the Tahiti video when a white backgroud is displayed as shown in the following screen capture. Is there any reason why you did use a discrete mask instead of an anti-aliased one ?

As I said, there's no mask being used. Anti-aliasing works on OpenGLES 2.0 devices where we have access to vertex and fragment shaders, but at this time we are using only OpenGLES 1.1. There's a lot of research being done to get the best image quality possible out of the lenticular screen on these (and desktop) devices, and I imagine the next version of 3DeeCentral will take advantage of vertex and fragment shaders to produce some impressive results.

moire.jpg

I guess you did use a mask similar to this one :

blackwhite.png

instead of one looking like this one, which is not perfect because I created it manually by copy&paste, but does not produce diagonal moiré :

blackwhite480aa.png

Another question I had is why you didn't create any game for the 3DeeSlide as was done for the 3DeeShell. I thought you were using an off-line interlacing that was not possible to render in a game in real-time, but since you've confirmed this to not be the case, I wonder why you didn't adapt the previous game (a space shooter IIRC) for the new lenticular screen.

Yeah that one was before we had gotten real-time interlacing working on the iDevices We just didn't update the game because it would be too much work and it just wasn't very good...we considered making a sequel though lol.

I've studied the 3DeeSlide lenticular lens quite in deep and I don't understand why you didn't implement multi-view either, since it's slanted (33.69° or atan(2/3) more precisely) and it covers approx. 4.62 horizontal pixels. My first tests seem to indicate that it should be possible to do it, why didn't you take advantage of this, is it because the resolution in 3D would be too low ?

Multiview works, but it's not that great. Not with the pre-retina devices anyway. I'm not sure what it looks like on the new ones, I imagine it works better though.

I'm still trying to obtain the best possible image quality and I'm currently stuck in the anti-aliasing techniques you seem to have implemented correctly on your side. I've read a lot of papers about this subject (Konrad, Agniel, Ruijters, etc.) but they're quite complicated and I lack the mathematical background to implement what they describe. Did you use the same techniques as the ones presented in these papers or did you use another solution ?

Thank you, and btw great job on your post-processing implementation.

Magic

[jgh]

I should add one thing - There was a piece of software available that allowed you to interlace your own content for 3DeeShell/Slide....I guess it's no longer around [I can't find it anyhow]. I also made an SDK at one point for game developers, but that needs to be updated... I'm sure that stuff will become available again at some point for the 3DeeSlide.

[Fredz]

Thanks for your answers, it's been very informative.

Pre-interlaced content is what we did for some of the 3DeeShell content before we properly figured out real-time interlacing on the iDevices. There is no mask used as you are suggesting, and lenticular interlacing is more complex for LCD screens than it is for print. The down-side to pre-interlaced content is because of the interlacing, it doesn't compress very well. So even short video ends up becoming quite huge. So it's a trade-off. Do we abandon lower-end devices, or do we force their owners to download massive pieces of content?

Ah, I didn't think about video compression, your approach sounds logical then. I used the mask only for videos in fact, for images I used an algorithm for interlacing, but the result is the same since I don't do pre-filtering for now.

As I said, there's no mask being used. Anti-aliasing works on OpenGLES 2.0 devices where we have access to vertex and fragment shaders, but at this time we are using only OpenGLES 1.1.

I wasn't talking about anti-aliasing filters for the content but only for the interlacing, that's what gives the moiré effect. Maybe you didn't consider it to be that annoying.

There's a lot of research being done to get the best image quality possible out of the lenticular screen on these (and desktop) devices, and I imagine the next version of 3DeeCentral will take advantage of vertex and fragment shaders to produce some impressive results.

Ah, so it seems you didn't use the same techniques than the ones presented in the research papers I talked about (based on Voronoi cells). It gives me hope to find an easier solution now.

Yeah that one was before we had gotten real-time interlacing working on the iDevices We just didn't update the game because it would be too much work and it just wasn't very good...we considered making a sequel though lol.

That would have been a good idea, especially considering the success the Nintendo 3DS is going to be in the following weeks.

Multiview works, but it's not that great. Not with the pre-retina devices anyway. I'm not sure what it looks like on the new ones, I imagine it works better though.

I'll try to create some multiview content to test the principle, but unfortunately I don't have an iPhone 4 yet.

Magic

Heh, that was my only technical question and I feared you would make an answer like this. I'll keep on searching for a good solution then...

[Fredz]

Yes, the autostereoscopic screens of both the Nintendo 3DS and the Fujifilm W3 should be better since the parallax barrier is incorporated into the screen, but the 3DeeSlide is still an interesting device. I'd like to see all of them with my own eyes to compare quality.

The fact that Spatial View is using a lenticular screen is quite interesting too since it doesn't halves the light output like parallax barriers and allows for multiple point of views and motion parallax. I guess it should also reduce moiré effects and I'd be curious to see how autostereo screens based on vertical barriers fare in this regard.

When the 3DeeSlide will be available for the iPhone 4 and with the front-camera for head-tracking, I'm not sure it won't be able to compete with the 3DS and the W3.

[option11]

Thanks for this great review and playing around with it. I bought 3Deeslide a while ago, hoping I'd be able convert my own 3D content for the Ipod (3DeeCentral is a bit of a joke at this stage I have to say).

I experimented a bit more and eventually wrote an iPod application in which I can play around with the mask properties (width of the lines, distance, angle...). After a fair bit of tweaking I ended up with the mask below, which seems to work perfectly for me.

I converted some 3D trailer with AVIsynth, this mask and converted them with iPodMe and the results were stunning. i didn't get any ghosting or moiree. I now want to figure out a quick way of converting side-by-side 3d images into 3deeslide compatible ones.

[Fredz]

I wrote a small application with Lazarus to convert stereo images to a format viewable with the 3DeeSlide. I'll post it if I remember when I'm back from holidays.

Concerning the mask it should be possible to have a better quality by using an anti-aliasing algorithm like the one proposed by Atanas Boev in one of his research papers, but I didn't find the time to actually implement this yet. His paper can't be found on the Web but if you ask him he'll send it to you. It's quite heavy on maths though...

[insertnick]

Concerning the mask it should be possible to have a better quality by using an anti-aliasing algorithm like the one proposed by Atanas Boev in one of his research papers, but I didn't find the time to actually implement this yet. His paper can't be found on the Web but if you ask him he'll send it to you. It's quite heavy on maths though...

Do you mean this one?

[Fredz]

Yes, it looks like this one, I didn't find it on the Web at the time I was looking for it, nice.

[insertnick]

I wrote a small application with Lazarus to convert stereo images to a format viewable with the 3DeeSlide. I'll post it if I remember when I'm back from holidays.

Back from holidays Fredz? :)