|
|
| pookien |
| Posted: Dec 13 2013, 04:39 PM |
 |
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
Infognition has just released a new version of its Super Resolution plugin for VirtualDub which upsizes video using information accumulated from neighbour frames (by using motion search and fusion of upscaled and compensated frames).
Unlike old version which only worked in RGB32, the new version natively supports YV12, YUY2 and RGB24 & RGB32. Speed increased 1.5 - 3 times, which made SD to HD conversion possible in real time on multicore boxes.
http://www.infognition.com/super_resolution_vdf/
AviSynth plugin is to be released soon too.
Those who have license for Video Enhancer can get a license for this plugin for free and vice versa. |
|
| evropej |
| Posted: Dec 16 2013, 02:42 AM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
virtualdub does a better job than this software. Went this route before, its easy with MSU plugins involved.
Here is a sample, 300% upscale
http://www.evropej.com/magic.jpg
and the trial version comparison
http://www.evropej.com/fail.jpg
which one would you use lol? |
|
| pookien |
| Posted: Dec 16 2013, 04:09 AM |
|
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
You're comparing very different things. Why don't compare with Need For Speed or MS Excel then? 
This plugin is only a resize. Not deblocking, not denoising. Give it a blocky and noisy image - get a noisy and blocky result.
The only thing it does is resizing, and the only thing it excels in is upsizing, providing more details in high frequencies.
So if you want to make a comparison, turn on all the other filters of your chain and just replace "resize (Lanczos3)" with SR. Only then it will make sense.
Here's a fair example. Doing just a 3x resize from 480x270 to 1440x810:
http://stuff.thedeemon.com/bjorn-3x-lan.png (resize Lanczos3)
http://stuff.thedeemon.com/bjorn-3x-sr2.png (SR in two steps: 200% then 150%)
http://stuff.thedeemon.com/bjorn-org.png (original frame)
from this video:
http://data.infognition.com/sr_samples/1x/bjorn.avi (18 MB)
Just open the pics in two tabs and switch back and forward to see the difference. |
|
| evropej |
| Posted: Dec 16 2013, 08:33 PM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
Thanks for posting the comparison but I fail to see your point.
Why should I pay $39.95 when I can do a better job with free software?
As for your example, you re-sized twice and only once with vdub.
Also, there is sharpening applied to the image, whether you call it up-scaling, up-sizing, or what ever you want.
The plugin also uses anti-aliasing I am assuming or some form of de-blocking.
I add de-blocking on mine because it necessary with most low resolution video.
The rest is smoothing and sharpening.
You can use what you like at the end of the day, I like free and I like good results.
I could not reproduce your results with their trial version.
PS I would die laughing if the guy actually has internal vdub filters just overlaid with a different gui lol. I dont have free time now but I think it would be worth someone taking a look at the assembly code lol. |
|
| raffriff42 |
| Posted: Dec 16 2013, 11:12 PM |
|
|
Advanced Member
Group: Members
Posts: 384
Member No.: 35081
Joined: 25-June 12
|
As stated in the OP,
"upsizes video using information accumulated from neighbour frames (by using motion search and fusion of upscaled and compensated frames)."
Works best with slowly moving complex textures (easily tracked with motion compensation)
Has trouble with aliased sources or where motion compensation fails (falls back on over-sharpening)
I can duplicate the results using the AVI source, but not using the still frame source.
Posted screen grabs appear to represent the absolute best case. Some frames do not look as clean as the one posted.
There is no (free) VirtualDub or Avisynth equivalent that I'm aware of. Still looking... |
|
| pookien |
| Posted: Dec 17 2013, 01:41 AM |
|
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
| QUOTE | | when I can do a better job with free software? |
Go on and reproduce the example above with free software. You're free to use any combination of free filters, you won't get the same level of details. SR was applied twice because each instance doubles the resolution inside, so it's optimal for 300%. But with Lanczos3 it's not the case, so applying Lanczos or any other intra-frame resize twice will not help.
| QUOTE | I add de-blocking on mine because it necessary with most low resolution video.
The rest is smoothing and sharpening. |
Ok, but why don't you add the same de-blocking and smoothing filters on the right side of your comparison? Why do you think this SR plugin would replace all of them? It never said it would, it's only a resize.
The idea is simple: with free plugins you can indeed get great results. With those plugins plus SR (used instead of internal resize) you can get even better.
| QUOTE | | I could not reproduce your results with their trial version. |
Why? Just open the file, apply the SR filter and go to frame 50 (which was used here).
If you just look at the first frame, it would be just intra-frame resize, you need to accumulate some info by processing a sequence of frames.
| QUOTE | | if the guy actually has internal vdub filters |
then it would be possible to get the same results using only those filters. But it's not. |
|
| evropej |
| Posted: Dec 17 2013, 02:16 AM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
Did you see the example I posted? I already have a process for upscaling and works great.
If I have enough time, I might do the same with your video. At the end of the day, there is no magic involved.
PS I can give you my video to upscale but the software will fail as shown. Interpolation is just part of upscaling. |
|
| pookien |
| Posted: Dec 17 2013, 02:54 AM |
|
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
Yes, I've seen your example and explained why it is incorrect.
There is no sense in comparing deblock+resize+sharpen+smooth vs. just another_resize.
Correct comparison would be deblock+resize+sharpen+smooth vs. deblock+another_resize+sharpen+smooth.
|
|
| evropej |
| Posted: Dec 17 2013, 05:57 AM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
My process works for resizing, how is it not correct? Because I use more filters lol?
In fact, that filter is insufficient for most resizing. |
|
| pookien |
| Posted: Dec 17 2013, 10:58 AM |
|
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
Your process is ok (probably might be improved by using SR plugin instead of Lanczos). Your comparison above is what's not correct.
|
|
| evropej |
| Posted: Dec 18 2013, 02:29 AM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
The only time the filter produces better results is with video which has slow motion and straight lines. Any other time, I can produce better results with just resizing. So, if you have a video which lots of straight lines, slow motion and you are forced to use a single filter, this is your filter lol. To me, this is not 40 dollars worth at all.
There are photoshop filters which do a lot more for the same price, I dont see the justification here. I wont bother testing the upscaling features in a tv or dvd players since the results with free filters are adequate. I dont see the justification.
If it makes you happy, buy it. |
|
| pookien |
| Posted: Dec 18 2013, 04:35 AM |
|
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
| QUOTE | | The only time the filter produces better results is with video which has slow motion and straight lines. Any other time, I can produce better results with just resizing. |
These claims are unjustified. You failed to show a single case where "just resizing" produces better results.
| QUOTE | | So, if you have a video which lots of straight lines, slow motion and you are forced to use a single filter, this is your filter lol. |
Why do you keep insisting on a single filter? Why on Earth do you think this plugin should replace all of them? |
|
| evropej |
| Posted: Dec 18 2013, 05:31 AM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
Here you go, just lanczos re-size
http://www.evropej.com/lol_lanczos.jpg
http://www.evropej.com/lol_thealmightysupe...rresolution.jpg
As you can see clearly from this video, the over-sharpening by the filter causes the artifacts to become worse and making the up-scaling look horrible.
Just lanczos up-scaling looks much better.
Like I said, you like it, use it but don't tell me its better for up-scaling because it does some sharpening while up-scaling at the same time.
It only works well with perfectly shot movies which have no noise, no blocking, no curves, no fast motion -period. |
|
| pookien |
| Posted: Dec 18 2013, 09:27 AM |
|
|
Advanced Member
Group: Members
Posts: 39
Member No.: 24784
Joined: 30-December 08
|
Thanks. What was the output resolution here, 13x the original? 
I agree, on low bitrate video with many artifacts and noise it looks worse. |
|
| jpsdr |
| Posted: Dec 18 2013, 06:25 PM |
|
|
Advanced Member
Group: Members
Posts: 335
Member No.: 20490
Joined: 23-December 06
|
I've been very interested by the potential result, so, i've decided to make tests.
My purpose is to upscale DVD, so from 480p to 720p.
The video is noisy, so, i've first denoised it, creating my orignal test source.
You can get a little part here : http://jpsdr.free.fr/XBMC/Test_Original.rar
So, using this video as source, i've upscaled it using two ways :
I've first upscale with NNEDI3, as i'm used to. I'm using the last version produced by the author, version you can find somewhere in the nnedi3 thread on doom9, version where author said he was using ABS(x-x') for distance of convergence criteria instead of (x-x')², which according him produced better results on "good part" (result less good on "bad part", but as they where allready bad...).
Code used :
| CODE |
AVISource("Ken - xxx.avi",False,"YV12")
SetPlanarLegacyAlignment(True)
nnedi3_rpow2(rfactor=2,cshift="Spline36Resize",fwidth=960,fheight=720,nsize=0,nns=3,qual=2)
|
Result here : http://jpsdr.free.fr/XBMC/Test_NNEDI3.rar
I've after uspcaled the source with SR2. For this, i've used 2 instances of the filter in VDub. A first instance with an upscale of x2. A second instance with a downscale to 960x720.
Result here : http://jpsdr.free.fr/XBMC/Test_SR2.rar
Video are encoded with UT Video Codec v13.3
Well... According your tests you're supposed to be better than NNEDI3, but... i don't see any differences...
Either my video is not the kind which can benefit of SR2, either, i'm doing something wrong...
Edit : In fact, after research, i'm just using v0.9.4 of nnedi3. |
|
