fauxD: 2D to stereo-3D conversion in real-time - Page 2

Fizick · 7th January 2009, 19:08

sorry for function names confusing

eslave · 8th January 2009, 01:58

Quote:

Originally Posted by Fizick

sorry for function names confusing

Don't worry about that, its my own fault for living in the past.

I really should have taken a closer look at 2.31 before my very first post. I still don't know what my 'hang-on-exit' issue was, but it was most assuredly my fault as well.

(Obviously I never called any MVtools functions.)

Unless something goes wrong I'll have a new version of fauxD posted later this evening. (Compatible with the current version of MVTools, 2.31) I still need to do some verification testing and fix the docs.

-eslave

setarip_old · 8th January 2009, 02:34

@eslave

May I ask if you've made any progress regarding converting this to a VirtualDub filter?

morsa · 8th January 2009, 09:09

I found one problem in this implementation of the Pulfrich Effect

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

The effect should only be applied on lateral motions, and from what I saw, this script/plugin doesn't take that into account.Vertical panning/motion also has the effect on it which is not a good thing.
Is there any possibility of using just displacements on the X axis?

eslave · 8th January 2009, 13:33

Well, that took far more testing and head-scratching than I had anticipated. Now with 16 different ways to call fauxD (together with its AVS doppleganger) its very easy to overlook something. I've added look-ahead for scene-changes, an 'always-on-top' checkbox, and fixed the 'flat-on-seek' issue rfmmars pointed-out. I've also removed all the 'V's from the MVTools function names for compatibility with Fizick's current version. (Thanks leeperry, mikeytown2, and rfmmars for repeatedly trying to point that out to me.)

Look-ahead will produce 2 duplicate stereo frames in sequence. My alternative was to use reverse stereo on the first frame, thereby furnishing a sense that 'every frame is different', but frequent false-positives caused eye-strain.

@morsa
Speaking of eye-strain, which I make little mention of in the doc, I know pulfrich should only be applied to purely horizontal video (with the camera moving at a prescribed rate), but I am not in control of the camera motions. True, I make no exceptions for vertical motion, I only care about pan direction ('displacements on the X axis'). Yes, this is an oversimplifcation with respect to pulfrich, but its the best I can do in realtime. I simply fuse 2 adjacent frames as stereo depending on pan direction. I wrote the ZeroDepan function hoping to apply some stability (vertical, rotation, zoom) to the source clip while preserving horizontal motion, but I haven't made any real effort to employ such a technique yet. I refer to scenes with strong vertical motion, or anything else that causes severe eye-strain, as "pulfrich-unfriendly". If you really want to see some depth in a video with only vertical motion then turn the video 90 degrees before processing. (Of course you'll need to watch the final output at the same orientation.)

@setarip_old
I've updated my earlier post. I still don't have a time-frame for a vdub port at the moment. Sorry.

@Shinobu
Oh my! Please don't get your expectations too high. This is purely a 'trick' as far as 3D-freaks are concerned. Also, if you happen to have 2 projectors maybe you can try polarized glasses.

I really appreciate all the help and testing that all of you have done, but I'm nearing the end of this current 'fit of creativity'. I'm not stopping development, but I want to actually enjoy fauxD for a little while before breaking it again.

Also, my new toy has just arrived and I'll probably be playing with that for at least a few days. So don't be concerned if I suddenly get a little 'quiet' around here. However, if I do notice that something serious is broken I'll try to correct it asap.

-eslave

Shinobu · 8th January 2009, 18:15

i don't expect to high, i'm just interested in testing this "trick".
as for the 2 projectors, i think i'll get a 3-dlp 1080p (when i'll have the money) instead of buying another 720p just to test 3D ^^.

by the way, i have an old 22" crt display, a 24" lcd and the 1-dlp projector, so i'll tell you witch one is the best renderer for your plugin (i'm going to test it with film and animation bdr).

morsa · 19th January 2009, 18:21

so, is there any way of not applying the effect on/as vertical displacements, even if it is not realtime ?

eslave · 26th January 2009, 05:41

Morsa, I'm uncertain what criteria to use for not applying pulfrich. Any vertical motion? Vertical motion exceeding horizontal motion? Some arbitrary combination of both? You could use GetTilt together with ScriptClip or ConditionalFilter to force the flat image to be returned instead. Remember, the analysis clip was vertically scaled to 1/2 or 1/4 height by default, and the values for vertical displacement will reflect this.

Maybe add a line to fauxDit() like this (only the blue lines are new):

Code:

...
out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scene_change_threshold), false )     #faster than RGB?
out = ConditionalFilter(out, out, flat, "abs(GetTilt(depan)*4)", "<", "1.0", false )     #skip vertical motion? (depan clip was scaled)
out = (stereo_mode == 3) ? out.Tweak(Bright=30) : out       #brighten final output (for shutterglasses)  AUTO
...

or:

Code:

...
out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scene_change_threshold), false )     #faster than RGB?
out = ConditionalFilter(out, out, flat, "abs(GetTilt(depan)*4)", "<", "abs(GetPan(depan))", false )     #skip vertical motion if greater than horizontal?
out = (stereo_mode == 3) ? out.Tweak(Bright=30) : out       #brighten final output (for shutterglasses)  AUTO
...

(Oops, I completely misread your post the first time, I thought you wanted fauxD to process vertical motion in some brilliant fashion.)

I'm certain there's a way to process vertical motion, and thats one of many future improvements I would like to add to fauxD, eventually. The current version of fauxD completely disregards vertical displacement. It would probably require real math and actual number crunching to produce such an effect, and I doubt I could do it in realtime.
My primary concern for this version was realtime processing.

However, MVTools will almost always report some horizontal motion, even during still scenes. I often think I see some depth (when there is no real horizontal motion) due to mild vertical motion at times, as long as the difference is only a few pixels. This is likely due to wishful thinking on my part, or chromatic disparity.

@Shinobu
I don't know if HD can be done in realtime, unless you have a ridiculously fast cpu. (I haven't even bothered trying on my system.) Be sure to let me know if you need any of that 'old' hardware tested.

-eslave

Also, another product, which I thought had been discontinued, appears to have been rebranded. '3DPlus' can be purchased and downloaded from the eDimensional website (in the 'Stunning 3D' section). There is no trial/demo available. You can find a critical review of the older versions here.

AnnaFan777 · 30th January 2009, 19:04

I suppose there's a way to do it vice versa, using mvtools
to convert this image into 2 separate full color pics.

(maybe use only edges for analysis, since (SAD) would be
way off from R and G&B )

need advice. example..

if succeed, we can convert 3d video to 2 full color vids,
heck we can then use lcd glasses to view 3d vid in full color...

eslave · 1st February 2009, 11:44

AnnaFan777, sorry to burst your bubble, but unfortunately with anaglyph too much information is lost for simple recovery of stereo-pairs. Also MVTools cannot do much with a still image (it never moves). I looked into this a few months ago, and it seems my best realtime option (with a still image or video) is to separate one of the color channels as the 'other' stereo image and then treat both images as some kind of greyscale (unless I want to attempt re-coloring things). I did find one person out there who actively rebuilds movies into various stereo-pair formats by using the original movie together with its anaglyph version and then SADing together anything that didn't fit right the first time (occlusions, etc).
Links: UNDERGROUND 3D CINEMA, Discussion1, Discussion2 (DeAnaglyph_Site1, Site2), Anaglyph to Stereo-Pairs

On another note I've quickly written an additional function for those of you with the 'new' Yellow(Amber)/Blue 3D glasses (for the superbowl commercial and other promotions). I haven't finished tweaking the brightness issues, but it does work. (I left the commented lines for any testers out there.)

Copy only the blue lines into fauxD.avsi as shown:

Code:

 ...function AnaglyphRedCyan(){}...
#///////////////////////////////////////////////////////////////////
# Create YB anaglyph from 2 clips. Used as a 'mode' function by FauxD()
function AnaglyphYellowBlue(clip left, clip right)
{
# Yellow (Amber) filter over left eye, Blue filter over right eye
# Left eye sees 'color' (RG)? Right eye sees 'depth' (B)?

#some believe creating a new monochrome 'channel' from greysale makes a better image (slightly brighter)
#right = right.greyscale(matrix="Rec601")
#right = right.greyscale(matrix="Average")

#boost blue channel for that dark filter.  (It may need more.  I haven't tested this because I don't have the glasses.)
#right = right.ConvertToYV12(matrix="PC.601").tweak(bright=30, coring=false).ConvertToRGB32(matrix="PC.601")    #brighten?
#right = right.ConvertToYV12(matrix="PC.601").tweak(cont=1.1, coring=false).ConvertToRGB32(matrix="PC.601")     #intensify contrast?
right = right.ConvertToYV12(matrix="PC.601").tweak(bright=12.5, sat=1.25, coring=false).ConvertToRGB32(matrix="PC.601")
right = right.Levels(0, 1.05, 255, 0, 255, coring=false)    #increase gamma?

#CopyChannel(RGtarget, BLUEsource)
return CopyChannel(left, right, 0)      #use plugin (should be faster?), blue is channel offset 0 (BGRA)

return  MergeRGB(left.ShowRed, left.ShowGreen, right.ShowBlue)      #AVS method
}
 ...function InterleavedShutter(){}...

 ...within function fauxDit()...
global reverse_stereo = false   #for the sake of completeness, affects AVS functions only
global avs_mode = (stereo_mode == 1) ? "OverUnder" : (stereo_mode == 2) ? "SideBySide"  :  (stereo_mode == 3) ? "InterleavedShutter" : (stereo_mode == 4) ? "AnaglyphYellowBlue" : "AnaglyphRedCyan"  # convert integer to string (function name) for fauxD/ScriptClip
use_plugin = (stereo_mode == 3) ? false : use_plugin      #'native' shutterglasses are only supported through the AVS function
 ...

and in your script call fauxDit() with mode=4 and plugin=false like this:

Code:

fauxDit(someclip, 4, 40, false)

-eslave

AnnaFan777 · 1st February 2009, 15:50

Quote:

Originally Posted by eslave

AnnaFan777, sorry to burst your bubble, but unfortunately with anaglyph too much information is lost for simple recovery of stereo-pairs. Also MVTools cannot do much with a still image (it never moves). I looked into this a few months ago, and it seems my best realtime option (with a still image or video) is to separate one of the color channels as the 'other' stereo image and then treat both images as some kind of greyscale (unless I want to attempt re-coloring things). I did find one person out there who actively rebuilds movies into various stereo-pair formats by using the original movie together with its anaglyph version and then SADing together anything that didn't fit right the first time (occlusions, etc).
Links: UNDERGROUND 3D CINEMA, Discussion1, Discussion2 (DeAnaglyph_Site1, Site2), Anaglyph to Stereo-Pairs

Maybe I didn't make it clear, Here's what i thought:

The still 3D pic still consists of two separate images, RED one and CYAN one.

1. create red's edge image A and cyan's edge image B (as in B&W comic book)
2. motion estimation -> from A TO B and vise versa. (it won't be perfect i know)
3. overlay original red image with motion compensated Cyan image, now you get 1 full color image
4. overlay original cyan image with motion compensated Red image, now you get the other full color image.

I'm a bit lazy and out of time, so if anyone's interested, please ...

.

eslave · 10th February 2009, 01:50

AnnaFan777, I believe the technique you're describing is pretty-much how the other anaglyph-to-stereo-pair methods are accomplished - except replace 'motion estimation' with 'edge-matching' or 'feature-tracking'. This is unlike anything fauxD presently does. I think it would require per-block or per-line matching methods, and I probably wouldn't be doing that in real-time. I'm reasonably certain there are AVIsynth tools capable of doing what you're talking about, but that would almost constitute another pet project in my opinion.

Attempting to use fauxD in its present form to de-anaglyph would likely produce bizarre results (or at least results more bizarre than normal). You could try playing with Shift() for some crude testing, but the red pixels in each original anaglyph frame will be shifted by varying amounts depending on the subject's depth:

Code:

clip = clip.converttorgb32
clip = MergeRGB(clip.ShowRed.Shift(8), clip.ShowGreen, clip.ShowBlue) #shift red channel 8 pixels (crude de-anaglyph)
fauxDit(clip, 1, 40, true) #return over-under stereo frames

Its recently come to my attention that fauxD may not produce any sound unless AC3Filter (and possibly other audio codecs) are installed. Also, if you're using fauxD.CMD and it doesn't launch your script (in your desired player) then you may need to remove the start command at the beginning of one line in the batch file.

-eslave

AnnaFan777 · 15th February 2009, 03:34

Quote:

Originally Posted by eslave

AnnaFan777, I believe the technique you're describing is pretty-much how the other anaglyph-to-stereo-pair methods are accomplished - except replace 'motion estimation' with 'edge-matching' or 'feature-tracking'. This is unlike anything fauxD presently does. I think it would require per-block or per-line matching methods, and I probably wouldn't be doing that in real-time. I'm reasonably certain there are AVIsynth tools capable of doing what you're talking about, but that would almost constitute another pet project in my opinion.

Thanks, do they have some special software for this?
i suppose they are not using avisynth?

eslave · 19th February 2009, 11:37

Actually, I still think AVISynth would be a good place to start.

I've never found anything about software deanaglyphing for video except that guy at UNDERGROUND 3D CINEMA, but he writes his own stuff (private). The links I posted are all the leads I have...until now.

Get your De-Anaglyph test script here. (I'd like to keep the discussions separate. I don't want this thread to become a catch-all for everything 3D.)

Your original idea was right. I hadn't foreseen the logical conclusion of all that edge/feature-tracking.

-eslave

Fizick · 20th April 2009, 19:50

Here is the message i have got from Jeff C:

Quote:

On WinXP SP2, after...
1) placing fauxD.avsi, fauxD.dll, fauxD.cmd, mvtools.dll, mvtools2.dll, fauxD_batchsample.avs and project mpgs in one directory,
2) changing fauxD.AVSI, fauxD.CMD and fauxD_ffdshowsample.avs or fauxD_sample.avs paths & filenames accordingly,
3) paste fauxD_ffdshowsample.avs lines into ffdshow's avisynth config dialog (and enable it's checkbox),
4) then drag and dropping fauxD_ffdshowsample.avs into mplayerc.exe (Media Player Classic),
this results in the dialog box appearing, the MPC window is resized correctly, but NO images appear just a blank screen which seems to be playing but only once even though set to repeat.
I've tried both mvtools v2.2.1(adding a "V" to "MV.." functions) and 2.3.1 as well as drag and drop movies on fauxD.CMD then use fauxD_batchsample.avs in MPC with the same results: the FauxD dialog then a blank playing screen.
Please help with any instructions/advise as soon as you've a moment as i'm excited as all who download FauxD to see it in action.
Thank you both so much for the great work,

I never tried FauxD.

godeep3d · 25th April 2009, 21:10

Hello Eslave,..
and thanks for the excellent programming, esp. fauxD.
On WinXP SP2, after...
(A)
1) placing fauxD.avsi, fauxD.dll, fauxD.cmd, mvtools.dll, mvtools2.dll, fauxD_batchsample.avs and project mpgs in one directory,
2) changing fauxD.AVSI, fauxD.CMD and fauxD_ffdshowsample.avs or fauxD_sample.avs paths & filenames accordingly,
3) pasting fauxD_ffdshowsample.avs lines into ffdshow's avisynth config dialog (and enable it's checkbox),
4) then drag and dropping fauxD_ffdshowsample.avs into mplayerc.exe (Media Player Classic),
...this results in a blank, uneditable dialog box appearing without Shift or Skew sliders, the MPC window is resized correctly, but NO images appear just a blank screen which seems to be playing but only once even though set to repeat.
I've tried both mvtools v2.2.1(adding a "V" to "MV.." functions) and v2.3.1 as well as drag and drop movies on fauxD.CMD then use fauxD_batchsample.avs in MPC with the same results: the FauxD blank dialog appears then a blank playing screen.

(B)
Then i used the complete fauxD tester program and pasted "fauxD_for_testers.avs" template in the project directory as "1b_fauxD_for_testers.avs" with all other files in common. I then edited its lines as follows:
EDITED : Import("fauxD.avsi")
EDITED : stereo_mode = 2 # 0 = Red-Cyan Anaglyph, 1 = Over-Under (Right Above, Left Below), 2 = Side-By-Side (Left, Right), 3 = shutterglasses (pure AVS, Left=Field1)
EDITED : f="C:\Documents and Settings\JeffCottle\Desktop\Best Realtime 2d to 3d with DDD Tridef Media Player using instant Depthmap\fauxD\All\ep2_clone_war_p640.mpg"
DISABLED: #out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scth), false ) #faster than RGB?
ENABLED: out = ConditionalFilter(flat, flat, stereo, "RGBDifferenceFromPrevious()", ">", "30", false ) #for RGB clips, slower?
ENABLED: out = out.ConvertToRGB24 #for TMPGEnc, VFApi, or any other application limited to 24-bit RGB (8-bits per color/channel)
This fixed the blank playback and resulted in...
1) the same blank,uneditable dialog but this time 2 full color frames finally displayed side-by-side BUT without stereoscopic or Pulfrich effect and..
2) the following message overlayed: "RGB Difference: RGB difference can only be calculated on RGB images ([Conditional Filter, Expresion 1], line 1)"

I found no reference to "RGBDifferenceFromPrevious" functions anywhere except fauxD_for_testers.avs so does FauxD support it?
How do i get a visible stereoscopic or Pulfrich effect with FauxD on a PC (WinXP SP2)?
Please help with any instructions/advise as soon as you've a moment as i'm excited as all who download FauxD to see it in action.
Thanks again for the great work,
Jeff Cottle: 780 425 4914 (please feel free to call)

Wilbert · 25th April 2009, 22:22

Quote:

I found no reference to "RGBDifferenceFromPrevious" functions anywhere except fauxD_for_testers.avs so does FauxD support it?

See documentation of ConditionalFilter. It expects a RGB clip. If you apply it to a YCbCr you need to enable

Code:

DISABLED: #out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scth), false ) #faster than RGB?

and disable to other one. Note, i never have used fauxD. So, i'm just guessing here ...

godeep3d · 26th April 2009, 02:56

Hello Wilbert,..
really appreciate your reply.
1)
You'd written "See documentation of ConditionalFilter. It expects a RGB clip."

But on my 2nd attempt i'd added and enabled the line...
flat = flat.ConvertRGB
...to send an RGB clip to the ConditionalFilter but recieved the same error. Is there a function to determine the colorspace of my ep2_clone_war_p640.mpg clip?

2)
You wrote:"If you apply it to a YCbCr you need to enable...ConditionalFilter(..YDifferenceFromPrevious..)"

This was enabled on my 1st attempt resulting in blank playback.

Gavino · 26th April 2009, 11:54

Quote:

Originally Posted by godeep3d

on my 2nd attempt i'd added and enabled the line...
flat = flat.ConvertRGB
...to send an RGB clip to the ConditionalFilter but recieved the same error. Is there a function to determine the colorspace of my ep2_clone_war_p640.mpg clip?

The colorspace of your source clip is irrelevant here.
For the changed ConditionalFilter line to work, both 'flat' and 'stereo' need to be RGB.
So your "flat = flat.ConvertRGB" needs to come before the line
stereo = stereo.MatchPixelFormat(flat)

[I don't really know anything about fauxD either, so this is just based on looking at this section of the code.]

godeep3d · 26th April 2009, 18:57

Hello Gavino,..

[QUOTE=Gavino;1278335]..So your "flat = flat.ConvertRGB" needs to come before the line
stereo = stereo.MatchPixelFormat(flat)

..as seen in the code segment below my RGB line did preceed the stereo line but still resulted in the RGB Difference error message. I've attached my project 1b_fauxD_for_testers.avs in Jeffs_fauxD_for_testers.zip for you or anyone with a little time to diagnose the problem. And the project movie can be downloaded from: ftp://ftp.fh-schmalkalden.de/pub/mul...e_war_p640.mpg

(EXAMPLE CODE SEGMENT):
#flat = flat.ConvertToYV12 #for YDifferenceFromPrevious
flat = flat.ConvertRGB #(MY METHOD)
stereo = stereo.MatchPixelFormat(flat)
out = ConditionalFilter(flat, flat, stereo, "RGBDifferenceFromPrevious()", ">", "30", false ) #for RGB clips, slower?
out = out.ConvertToRGB24 #for TMPGEnc, VFApi, or any other application limited to 24-bit RGB (8-bits per color/channel)

Thank you all for every suggestion and any help,
Jeff

7th January 2009, 19:08	#21 \| Link
Fizick AviSynth plugger Join Date: Nov 2003 Location: Russia Posts: 2,183	sorry for function names confusing __________________ My Avisynth plugins are now at http://avisynth.org.ru and mirror at http://avisynth.nl/users/fizick I usually do not provide a technical support in private messages.

8th January 2009, 13:33	#25 \| Link
eslave Registered User Join Date: Dec 2008 Location: reviving a dead thread Posts: 31	Version 0.12 Posted Well, that took far more testing and head-scratching than I had anticipated. Now with 16 different ways to call fauxD (together with its AVS doppleganger) its very easy to overlook something. I've added look-ahead for scene-changes, an 'always-on-top' checkbox, and fixed the 'flat-on-seek' issue rfmmars pointed-out. I've also removed all the 'V's from the MVTools function names for compatibility with Fizick's current version. (Thanks leeperry, mikeytown2, and rfmmars for repeatedly trying to point that out to me.) Look-ahead will produce 2 duplicate stereo frames in sequence. My alternative was to use reverse stereo on the first frame, thereby furnishing a sense that 'every frame is different', but frequent false-positives caused eye-strain. @morsa Speaking of eye-strain, which I make little mention of in the doc, I know pulfrich should only be applied to purely horizontal video (with the camera moving at a prescribed rate), but I am not in control of the camera motions. True, I make no exceptions for vertical motion, I only care about pan direction ('displacements on the X axis'). Yes, this is an oversimplifcation with respect to pulfrich, but its the best I can do in realtime. I simply fuse 2 adjacent frames as stereo depending on pan direction. I wrote the ZeroDepan function hoping to apply some stability (vertical, rotation, zoom) to the source clip while preserving horizontal motion, but I haven't made any real effort to employ such a technique yet. I refer to scenes with strong vertical motion, or anything else that causes severe eye-strain, as "pulfrich-unfriendly". If you really want to see some depth in a video with only vertical motion then turn the video 90 degrees before processing. (Of course you'll need to watch the final output at the same orientation.) @setarip_old I've updated my earlier post. I still don't have a time-frame for a vdub port at the moment. Sorry. @Shinobu Oh my! Please don't get your expectations too high. This is purely a 'trick' as far as 3D-freaks are concerned. Also, if you happen to have 2 projectors maybe you can try polarized glasses. I really appreciate all the help and testing that all of you have done, but I'm nearing the end of this current 'fit of creativity'. I'm not stopping development, but I want to actually enjoy fauxD for a little while before breaking it again. Also, my new toy has just arrived and I'll probably be playing with that for at least a few days. So don't be concerned if I suddenly get a little 'quiet' around here. However, if I do notice that something serious is broken I'll try to correct it asap. -eslave

26th January 2009, 05:41	#28 \| Link
eslave Registered User Join Date: Dec 2008 Location: reviving a dead thread Posts: 31	Morsa, I'm uncertain what criteria to use for not applying pulfrich. Any vertical motion? Vertical motion exceeding horizontal motion? Some arbitrary combination of both? You could use GetTilt together with ScriptClip or ConditionalFilter to force the flat image to be returned instead. Remember, the analysis clip was vertically scaled to 1/2 or 1/4 height by default, and the values for vertical displacement will reflect this. Maybe add a line to fauxDit() like this (only the blue lines are new): Code: ... out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scene_change_threshold), false ) #faster than RGB? out = ConditionalFilter(out, out, flat, "abs(GetTilt(depan)4)", "<", "1.0", false ) #skip vertical motion? (depan clip was scaled) out = (stereo_mode == 3) ? out.Tweak(Bright=30) : out #brighten final output (for shutterglasses) AUTO ... or: Code: ... out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scene_change_threshold), false ) #faster than RGB? out = ConditionalFilter(out, out, flat, "abs(GetTilt(depan)4)", "<", "abs(GetPan(depan))", false ) #skip vertical motion if greater than horizontal? out = (stereo_mode == 3) ? out.Tweak(Bright=30) : out #brighten final output (for shutterglasses) AUTO ... (Oops, I completely misread your post the first time, I thought you wanted fauxD to process vertical motion in some brilliant fashion.) I'm certain there's a way to process vertical motion, and thats one of many future improvements I would like to add to fauxD, eventually. The current version of fauxD completely disregards vertical displacement. It would probably require real math and actual number crunching to produce such an effect, and I doubt I could do it in realtime. My primary concern for this version was realtime processing. However, MVTools will almost always report some horizontal motion, even during still scenes. I often think I see some depth (when there is no real horizontal motion) due to mild vertical motion at times, as long as the difference is only a few pixels. This is likely due to wishful thinking on my part, or chromatic disparity. @Shinobu I don't know if HD can be done in realtime, unless you have a ridiculously fast cpu. (I haven't even bothered trying on my system.) Be sure to let me know if you need any of that 'old' hardware tested. -eslave Also, another product, which I thought had been discontinued, appears to have been rebranded. '3DPlus' can be purchased and downloaded from the eDimensional website (in the 'Stunning 3D' section). There is no trial/demo available. You can find a critical review of the older versions here. Last edited by eslave; 2nd May 2009 at 07:06. Reason: need glasses

30th January 2009, 19:04	#29 \| Link
AnnaFan777 Registered User Join Date: Apr 2008 Posts: 51	I suppose there's a way to do it vice versa, using mvtools to convert this image into 2 separate full color pics. (maybe use only edges for analysis, since (SAD) would be way off from R and G&B ) need advice. example.. if succeed, we can convert 3d video to 2 full color vids, heck we can then use lcd glasses to view 3d vid in full color... Last edited by AnnaFan777; 30th January 2009 at 19:14.

10th February 2009, 01:50	#32 \| Link
eslave Registered User Join Date: Dec 2008 Location: reviving a dead thread Posts: 31	AnnaFan777, I believe the technique you're describing is pretty-much how the other anaglyph-to-stereo-pair methods are accomplished - except replace 'motion estimation' with 'edge-matching' or 'feature-tracking'. This is unlike anything fauxD presently does. I think it would require per-block or per-line matching methods, and I probably wouldn't be doing that in real-time. I'm reasonably certain there are AVIsynth tools capable of doing what you're talking about, but that would almost constitute another pet project in my opinion. Attempting to use fauxD in its present form to de-anaglyph would likely produce bizarre results (or at least results more bizarre than normal). You could try playing with Shift() for some crude testing, but the red pixels in each original anaglyph frame will be shifted by varying amounts depending on the subject's depth: Code: clip = clip.converttorgb32 clip = MergeRGB(clip.ShowRed.Shift(8), clip.ShowGreen, clip.ShowBlue) #shift red channel 8 pixels (crude de-anaglyph) fauxDit(clip, 1, 40, true) #return over-under stereo frames Its recently come to my attention that fauxD may not produce any sound unless AC3Filter (and possibly other audio codecs) are installed. Also, if you're using *fauxD.CMD* and it doesn't launch your script (in your desired player) then you may need to remove the start command at the beginning of one line in the batch file. -eslave

8th January 2009, 02:34	#23 \| Link
setarip_old Registered User Join Date: Aug 2005 Posts: 16,267	@eslave May I ask if you've made any progress regarding converting this to a VirtualDub filter?

8th January 2009, 09:09	#24 \| Link
morsa the dumbest Join Date: Oct 2002 Location: Malvinas Posts: 494	I found one problem in this implementation of the Pulfrich Effect http://en.wikipedia.org/wiki/Pulfrich_effect The effect should only be applied on lateral motions, and from what I saw, this script/plugin doesn't take that into account.Vertical panning/motion also has the effect on it which is not a good thing. Is there any possibility of using just displacements on the X axis?

8th January 2009, 18:15	#26 \| Link
Shinobu Satsuki'S Softs Maker Join Date: Feb 2003 Posts: 192	i don't expect to high, i'm just interested in testing this "trick". as for the 2 projectors, i think i'll get a 3-dlp 1080p (when i'll have the money) instead of buying another 720p just to test 3D ^^. by the way, i have an old 22" crt display, a 24" lcd and the 1-dlp projector, so i'll tell you witch one is the best renderer for your plugin (i'm going to test it with film and animation bdr).

19th January 2009, 18:21	#27 \| Link
morsa the dumbest Join Date: Oct 2002 Location: Malvinas Posts: 494	so, is there any way of not applying the effect on/as vertical displacements, even if it is not realtime ?

19th February 2009, 11:37	#34 \| Link
eslave Registered User Join Date: Dec 2008 Location: reviving a dead thread Posts: 31	Actually, I still think AVISynth would be a good place to start. I've never found anything about software deanaglyphing for video except that guy at UNDERGROUND 3D CINEMA, but he writes his own stuff (private). The links I posted are all the leads I have...until now. Get your De-Anaglyph test script here. (I'd like to keep the discussions separate. I don't want this thread to become a catch-all for everything 3D.) Your original idea was right. I hadn't foreseen the logical conclusion of all that edge/feature-tracking. -eslave

25th April 2009, 21:10	#36 \| Link
godeep3d Registered User Join Date: Apr 2009 Posts: 3	fauxD 2D to stereo-3D conversion in real-time How-To Hello Eslave,.. and thanks for the excellent programming, esp. fauxD. On WinXP SP2, after... (A) 1) placing fauxD.avsi, fauxD.dll, fauxD.cmd, mvtools.dll, mvtools2.dll, fauxD_batchsample.avs and project mpgs in one directory, 2) changing fauxD.AVSI, fauxD.CMD and fauxD_ffdshowsample.avs or fauxD_sample.avs paths & filenames accordingly, 3) pasting fauxD_ffdshowsample.avs lines into ffdshow's avisynth config dialog (and enable it's checkbox), 4) then drag and dropping fauxD_ffdshowsample.avs into mplayerc.exe (Media Player Classic), ...this results in a blank, uneditable dialog box appearing without Shift or Skew sliders, the MPC window is resized correctly, but NO images appear just a blank screen which seems to be playing but only once even though set to repeat. I've tried both mvtools v2.2.1(adding a "V" to "MV.." functions) and v2.3.1 as well as drag and drop movies on fauxD.CMD then use fauxD_batchsample.avs in MPC with the same results: the FauxD blank dialog appears then a blank playing screen. (B) Then i used the complete fauxD tester program and pasted "fauxD_for_testers.avs" template in the project directory as "1b_fauxD_for_testers.avs" with all other files in common. I then edited its lines as follows: EDITED : Import("fauxD.avsi") EDITED : stereo_mode = 2 # 0 = Red-Cyan Anaglyph, 1 = Over-Under (Right Above, Left Below), 2 = Side-By-Side (Left, Right), 3 = shutterglasses (pure AVS, Left=Field1) EDITED : f="C:\Documents and Settings\JeffCottle\Desktop\Best Realtime 2d to 3d with DDD Tridef Media Player using instant Depthmap\fauxD\All\ep2_clone_war_p640.mpg" DISABLED: #out = ConditionalFilter(flat, flat, stereo, "YDifferenceFromPrevious()", ">", String(scth), false ) #faster than RGB? ENABLED: out = ConditionalFilter(flat, flat, stereo, "RGBDifferenceFromPrevious()", ">", "30", false ) #for RGB clips, slower? ENABLED: out = out.ConvertToRGB24 #for TMPGEnc, VFApi, or any other application limited to 24-bit RGB (8-bits per color/channel) This fixed the blank playback and resulted in... 1) the same blank,uneditable dialog but this time 2 full color frames finally displayed side-by-side BUT without stereoscopic or Pulfrich effect and.. 2) the following message overlayed: "RGB Difference: RGB difference can only be calculated on RGB images ([Conditional Filter, Expresion 1], line 1)" I found no reference to "RGBDifferenceFromPrevious" functions anywhere except fauxD_for_testers.avs so does FauxD support it? How do i get a visible stereoscopic or Pulfrich effect with FauxD on a PC (WinXP SP2)? Please help with any instructions/advise as soon as you've a moment as i'm excited as all who download FauxD to see it in action. Thanks again for the great work, Jeff Cottle: 780 425 4914 (please feel free to call)

26th April 2009, 02:56	#38 \| Link
godeep3d Registered User Join Date: Apr 2009 Posts: 3	fauxD 2D to stereo-3D conversion in real-time How-To Hello Wilbert,.. really appreciate your reply. 1) You'd written "See documentation of ConditionalFilter. It expects a RGB clip." But on my 2nd attempt i'd added and enabled the line... flat = flat.ConvertRGB ...to send an RGB clip to the ConditionalFilter but recieved the same error. Is there a function to determine the colorspace of my ep2_clone_war_p640.mpg clip? 2) You wrote:"If you apply it to a YCbCr you need to enable...ConditionalFilter(..YDifferenceFromPrevious..)" This was enabled on my 1st attempt resulting in blank playback.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode