The differences are not as big as in the first renders (SRv4 and below I think), but there are also fewer artifacts while you can clearly see that detail is recovered (for example Vader's arm in frame 8228).
Jagged lines are also reduced in comparison to earlier attempts.
So, is the Infognition software still in your workflow?
In marked contrast to the spam bot above me, I have found the solution. My files were too big, over the 4 GB limit of WAV files. I knew that there was a limit, I just thought that it will stand out much earlier, at the point of file creation.
Now I chopped the file down and had to fix them because of another problem with Premiere Pro:
http://mshg33.blogspot.de/2015/04/why-do-i-get-broken-wav-files-with.html
Yes, even downsampling (1080p -> SR 4k -> Spline64 1080p) a Blu-Ray does not show any noticable difference. So it's not only because of YouTube.
I tried this today when experimenting for a project, but with AOTC:
http://screenshotcomparison.com/comparison/123910
Best results are at the weapon's scope. Overall it introduces banding at gradients through more random noise, take a look at the edges of the first soldier's hat. Although AOTC and that frame might not be the ideal test subject.
Nice summary, thanks for writing it all down.
DrDre, that looks a bit too smooth for my eyes. For example the "corners" of the pillar in the first screenshot. And there is this weird pattern over all the walls that looks like the white is painted on with a brush.
mverta said:
Very rarely can I say with absolute certainty that a particular piece of damage is "original negative damage." Say it's one of those white spots we see in some shots, one that shows up on every print that's ever been struck (that we've SEEN). Should that be preserved? Even then, I can't say that white spot was on the original negative as it first came out of the camera after shooting. And if it wasn't, then it's not original negative, it's damage/dirt/dust. Even if it happened on the way from the camera mag to the developer, it's not original negative. It was shit floating in the air. And finding some clear definition for where in the process shit on the negative is good shit and where it's bad shit becomes random and insane and arbitrary. What, one hour after filming is good shit, but if it showed up two days later while striking IP's it's bad shit? It's impossible to say, I mean literally impossible. So in this case, I've decided to follow my own definition, which asks the question: Is it IN the scene, or ON the scene? If it's on the scene (physically on the emulsion of the negative), then it goes. Nuked from orbit; the only way to be sure. If it's something that's IN the scene (a crewmember in the shot) it stays.
Very well said, I like your philosophy on this.
By the way, will this be released to the public? The videos really got me interested.
Great work, both of you! In some scenes I prefer SRv5, in some LaserDisc Master's work.
@ Laserdisc Master
Do you plan to release a version with your MagicUp?
Sorry, I cannot give you the exact frame numbers as I do not have the NTSC GOUT. Preferrably somewhere near these two frames that we have already seen: this and this. And maybe another one that we have not seen, but two is already enough, thanks :)
Again it is stunning to see the comparison with bicubic resize, haven't seen it for a while ;)
Your latest test 122645 has even more detail but also more noise, and take a look at the gun's "sights" - the bead - it gets a bit deformed. This has happened in other results aswell, but here it is stronger. I would like to see other frames of MagicUpSR3 refined + GrainPlate, could you please post more maybe?
This is getting better and better, one might suspect you're using a real HD transfer, haha ;)
Regarding this comparison http://screenshotcomparison.com/comparison/122417 super resolution seems to have more detail which goes with more noise, but MU handles edges with aliasing and artifacts better (right edge of chinstrap seems to be a haloing artifacts, but I might be wrong).
_,,,^..^,,,_ said:
SuperResolution V3 Vs MagicUpPlus + GrainPlate2
http://screenshotcomparison.com/comparison/121901Spline64Resize + GrainPlate2 Vs MagicUpPlus + GrainPlate2
http://screenshotcomparison.com/comparison/121902Spline64Resize + GrainPlate2 Vs SuperResolution V3
http://screenshotcomparison.com/comparison/121903[...]
my thoughts: despite the fact SuperResolution V3 is (obviously) cleaner than "classic" upscale filters, it relies on artifacts... indeed, you could see that it distorts some shapes (eyes, last teeth etc); MagicUpPlus increments details while maintaining shapes... I added a grain plate to give the impression of further details.
Yes, MagicUpPlus really is better at keeping the shapes close to the Spline64Resize which enhancing the detail greatly. But the picture seems to be brightened up more by the process.
Did you improve it with the method DrDre used, different color spaces and reversing?
Thanks, I think that looks amazing.
Interesting project, I really like the colors and denoising in the third screenshot. Can't wait to see more! Could you maybe show one of the scene with the sandcrawler rolling over the dunes?
Yes, there are artifacts visible in the eyes.
Still those results are really impressive - wow!
What do you mean by "different orders"?
Ah, I see. So we could only benefit from the full potential if our material was completely dirt-free.
But seriously, I would take that better resolution even with artifacts.
I can barely make out any differences in your last comparison.
Could you show us some of the mentioned artifacts?
No, it is not. I guess unfortunately I have to find another source for this.
Thank you for your help.
Thank you for your answer, Jonno.
I forgot to mention that I already tried eac3to, it looked like it worked its way through the whole file but ended up giving me only the first 7 minutes.
Do you know how to fix a stream error?
Hello all,
I have a *.mkv file with a 5.1 DTS-HD Master Audio Stream. I extracted it and converted it to a multi-channel wave file. The result is a very slow audio (approx. 3 times slower) and still it is shown as 20 minutes shorter than the original audio.
So far I tried extracting/converting with TSMuxer, Avidemux, MKVToolnix, VLC and foobar2000. I either get a "wrong file format"/"unsupported compression" error in Premiere Pro when converting to *.dts or *.ts or a corrupted file like described above when converting to wav/FLAC. Maybe my MKV is broken, but I don't know how to fix it.
Can somebody help?
Thanks in advance
Which German audio can I use for this?
Wow, PaNup with MagicUp looks nice! But it also emphasizes the grain.
By the way, to simulate a scene change, I cut the clip so it starts from the given frame and used VideoEnhancer. Comparison with full 40-frame clip here:
http://screenshotcomparison.com/comparison/120395
Very subtle but you can see it better at her hairline. The most moving part, the mouth, has very few visible differences for me. But when zoomed in with Photoshop you can see it.
Note that real results depend on the algorithm on how well VideoEnhancer recognizes a scene change.
I think the VideoEnhancer only looks better than the script.
This is somewhat late, but you can add the chapters with MKVToolnix (mmg.exe):
Episode IV: http://textuploader.com/xl42
Episode V: http://textuploader.com/xl4p
I don't have any for Episode VI and there are not the appropriate ones on ChapterDB, only with differing runtimes.
"Same as source" and Constant Framerate is what I would choose when the input file has a constant framerate. Maybe see the Handbrake wiki for more info.
What exactly is your problem?