Has anybody run the gout v3 (gout with g-force script) into a dvdo iscan upscaler/processor, and if so, what were the results like and is it worthwhile creating distribution based on grabbing the iscan output ?
It's smarter to modify the script to output to the resolution you want. When I ran my modified version of G-Force's script I had it output to a 1:1 pixel ratio instead of anamorphic, without going through further steps of scaling.
[ Scanning stuff since 2015 ]
What does the 1:1 pixel ratio achieve vs. the anamorphic?
You can make it a normal AVI file.
[ Scanning stuff since 2015 ]
What would be interesting is what such a device would do with the raw GOUT. I'm not criticizing g-force, his script has been a huge step towards a quality improvement, but that script does involve heavy antialiasing and other things that eat up some of the vertical resolution.
I have a DVDO EDGE which does about the same processing as the ISCAN. Also have a Pioneer Elite Laser Disc player and JVC D-VHS VCR and can attest to what this processor is capable of achieving with low resolution sources. My capture device is a Hauppauge! HD PVR that is connected via USB to my computer. It can do 1080i and 720P and also can capture the raw Dolby Digital audio from LD and DVD's. I own the GOUT Trilogy and can do some sample captures if you guys are interested. I have Dark_Jedi's V3 DVD's also and can see how these look up-converted.
That would be great ( a few small comparison clips), if it's not too much trouble. I know there's upscaling projects on the go, but if the DVDO device gives superior results, and does it easily, then this could be applied to all sources and may be useful to people like Harmey in despecial-ing projects.
I don't own a ps3 but apparently (according to a website i just came across) it has an upscaler hidden in it's configuration settings somewhere. Has anyone checked out the results this gives? I assume the DVDO EDGE will kick it's **s. But while we're waiting for some screenshot and/or clips from the DVDO EDGE , it would be interesting to see the results the PS3 gives.
laserdisc said:
Interesting. I didn't know that. I'm still curious about the potential increase in quality of running gforce/gout through an iscan. It doesn't just do upscaling but also does general image enhancement apparently. I've seen some people mention iscans on this forum but in relation to laserdiscs. So regardless of whether it's avi or vob i'd love to know how much quality is improved with an iscan.
You won't ever get the best results for the GOUT using a single upscale method. Here's an example of the improvement in overall picture quality that can be achieved using one method...
Top frame is a somewhat stabalized GOUT (no antialising has been applied before upscale), the next two show two takes at an almost idental upscale method. The first is using a filter in avisynth (nnedi3), and the second is using a commercial filter (Video Enhancer). Both are resized using the same method (Spline36Resize), both have the same simple filters applied to the upscale. As you will see there is almost no difference between nnedi3 and VE; however while these filters work well on some shots, they work terrible on others. Garbage-in-garbage-out, both display problems handling the ghosting trails left by the DVNR and both also have trouble in some space scenes. Furthermore VE cannot be used as a frameserver (although you can use avisyth filters within the program itself) which makes nnedi3 far more attractive.
Larger view (full resolution for 720p):
[ Scanning stuff since 2015 ]
Best upscaling/antialiasing results I've seen so far is whatever dark_jedi used on his Blu V2 sample. I compared the sample against the output from a nnedi3 upscale, and it was much better.
GOUT NNEDI3
DJ Blu V2 Sample
Guidelines for post content and general behaviour: read announcement here
Max. allowable image sizes in signatures: reminder here
Moth3r said:
Best upscaling/antialiasing results I've seen so far is whatever dark_jedi used on his Blu V2 sample. I compared the sample against the output from a nnedi3 upscale, and it was much better.
GOUT NNEDI3
DJ Blu V2 Sample
I think his results are repetable, if not beatable in AVISynth. I only did a quick test where there was very little done to the image prior to upscaling. DJ's version has had substantial grain reduction and anti-alising applied before upscale and getting that right is definatly the key to improving the upscale result; however as I mentioned you still need to handle different types of scenes differently, espcially space scenes and scenes where the DVNR was very heavy.
Here is that scene with some grain removal and aa applied to the raw PAL gout (notice how the framing is slightly different, I would expect better success for NTSC especially in regards to anti-alising). NNEDI3 is then applied.
[ Scanning stuff since 2015 ]
RU.08 said:
DJ's version has had substantial grain reduction and anti-alising applied before upscale and getting that right is definatly the key to improving the upscale result; however as I mentioned you still need to handle different types of scenes differently, espcially space scenes and scenes where the DVNR was very heavy.
I wrote the script for V2 and there was no grain reduction at all in the script, and no antialiasing before the upscaling. The key for detail preservation vs jaggies reduction was to upscale before doing the processing.
RU.08 said:
Here is that scene with some grain removal and aa applied to the raw PAL gout (notice how the framing is slightly different, I would expect better success for NTSC especially in regards to anti-alising). NNEDI3 is then applied.
Yeah, the NTSC has more detail intact than the PAL. Weird.
In that picture you posted there is way too much detail being removed by the antialiasing though. (plus it's the PAL version)
I can tell you from experience that generally, EEDI2 gives a much better result when done in multiple passes with tweaked settings than NNEDI2 and NNEDI3 will ever do.
Especially if you use a plugin called QTGMC with EEDI2 as antialiaser:
Yeah, the NTSC has more detail intact than the PAL. Weird.
@Darth Editous: If they're upscaled NTSC there has to be some blurring involved because I have tried upscale scripts on both NTSC and PAL and it's like PAL is missing detail somehow. The PAL versions are also cropped differently.
And yes, PAL ROTJ is the exception and actually has more detail than the NTSC version.
Interesting. eedi2, qtgmc, turnleft, eedi2, turnright, spline36resize, LSFMod. I actually think I may have lost more detail this way, see what you think:
[ Scanning stuff since 2015 ]
What's the idea behind using qtgmc after the first eedi2?
DE
@RU.08: I hate to sound like a grumpy critic, but I get the feeling that you're almost trying to recreate the result me and DJ have gotten so far.
If you're just curious about our methods, I can tell you that we might reveal some more of it once the project blu V2 is finished. If you just wanna try and make your own preservation, maybe you should start an own thread.
You_Too said:
@RU.08: I hate to sound like a grumpy critic, but I get the feeling that you're almost trying to recreate the result me and DJ have gotten so far.
If you're just curious about our methods, I can tell you that we might reveal some more of it once the project blu V2 is finished. If you just wanna try and make your own preservation, maybe you should start an own thread.
You_Too,
RU.08's comments seem on-topic. I also kinda like what I'm seeing. Yeah, a bit of detail loss, but also much less sharpening artifacts than in your example.
Whether in this thread, or a new one, I hope he keeps going.
-G
All I've seen of project blu is what's in the forum and stuff Harmy used in his despecialisation. I personally don't know how I feel about using EEDI2 in a way that it isn't designed; by using it to stretch the picture horizontally it intorduces far more side-effects, for instance have a look at the background in your R2 pic, You_Too, where the stone patterns have been somewhat reconfigured by EEDI2 joining what it thinks are "lines". Even the scuff marks on R2D2 have also been reconfigured into a pattern thanks to EEDI2. I was wrong about the grain removal, it's been obscured by EEDI2.
In any case, as I said to begin with I believe the results are beatable. I need to aquire the NTSC disc to go much further. But for now yes I've removed the jaggies without using qtgmc (thank goodness because it's frightfully slow) or loosing much if any quality, here's a screenshot you can compare it to my earlier one from the same scene. For reference, I'm using EEDI2 once for deinterlization and then nnedi3 for scaling (although it makes little difference at this point if I use nnedi3 at all):
[ Scanning stuff since 2015 ]
@g-force: Sharpening or not, it's a matter of taste of course. Though it looks much more smooth when in motion.
@RU.08: I'm sorry if I misunderstood your intentions. Maybe we just have different preferences when it comes to upscaling.
You_Too said:
@g-force: Sharpening or not, it's a matter of taste of course. Though it looks much more smooth when in motion.
@RU.08: I'm sorry if I misunderstood your intentions. Maybe we just have different preferences when it comes to upscaling.
There are definitely some things I like better about You_Too's approach, but also some things I like better about RU.08's too.
I really like the line thinning and sharper edges of You_Too's example, but then there's ringing on things like R2's blue dome ring and utility arms that look like horizontal stripes.
And I like how RU.08's logic display on R2 looks like discreet points of light and less amorphous. Maybe you guys should get together on this, as if you really do have different preferences, you may come to a better compromise.
Either way, I hope both of you keep up the good work!
-G
You_Too said:
@RU.08: I'm sorry if I misunderstood your intentions. Maybe we just have different preferences when it comes to upscaling.
Perhaps. I believe EEDI2 will give the best results for (at least) most scenes in ANH, and I can't believe I let it slip me by until you mentioned it. But I certainy don't agree with using it to scale horizontally, a straight splineresize would work better then using EEDI2 horizontally, and now that you've mentioned it the "pattern" it introduces when used in both directions is pretty much unmissable. Anyway, I'm going to see how EEDI2 goes on Robocop Criterion now; that DVD has a lot of the same problems as ANH (DVNR and alising) and the fact that I'm not working with an upscale will mean I should get good results (except for robovision scenes, obviously), I'll make a new thread once I have a script looking good, and I'll come back to SW (ANH & ESB) once I have the NTSC version. :) Yes, I'll make my own thread when I feel I have a compentent script. :)
Darth Editous said:
What's the idea behind using qtgmc after the first eedi2?
DE
Good question! EEDI2 doubles the vertical resolution and each new line is an interpolation, but the way that EEDI2 works it tries to join "lines" like diagonal lines together (in the new lines), and that in itself very effectivly reduces the "jaggies". It is not designed to work horizontally, but you can force it to by using turnleft, etc. EEDI2 does not modify the original lines which let's say are now the "odd lines" and the new ones being the "even lines", it only creates new lines. qtgmc is an advanced deinterlacer, but a very slow one, and will reduce the jaggies in the image without resizing; so essentially using it is hitting the image with a second "deinterlacer". You wouldn't want to use it before EEDI2 because then it's "damaging" the original picture before resize; plus the fact that EEDI2 already does a lot of the work for you means you really only need a much lighter touch with a deinterlacer post-eedi2. And as I've already mentioned I don't think EEDI2 should be used to stretch the picture horizontally because it then "joins more lines together" and your image comes out looking like it is made up of crisscross strokes. And sorry for missing your question the first time. :)
[ Scanning stuff since 2015 ]