|
Quote:
|
Quote:
|
Quick update
1 Attachment(s)
Hey all,
The AIW 9000 came in on Monday. I popped that in and didn't change anything else about the set up, and it appears that the herringbone noise is gone, though I'm not 100% sure. Here is an excerpt of the beginning of Les Miz again(*). What say you? Separately - I'm having a bit of an issue with compression that I can't figure out. Not really related to this thread so I made a new thread for it here. (*) I put this tape in one of my old VCRs to see if it still worked, and it promptly got mangled and spit out. That's why you have the prominent white ziggies at some point. I won't be using this tape anymore for testing, I'm afraid... |
And rockovids - I definitely appreciate your advice here. I may still try with the chokes because I hate the sound of this AIW 9000 fan...
|
3 Attachment(s)
Bonus time!
I received my BJC-2 S-Video cables earlier this week. So how's this for a bonus - a comparison of an original 9800 cap (which I haven't posted here before), one with the 9000 using my original cables with ferrite cores on them, and one with the 9000 using a couple of YC-2s. Note that I think I touched the levels on the original 9800 cap when making it, so unfortunately the comparison may not be exact, but you could clearly see the difference between the two in terms of interference - which I don't think is completely eradicated with the 9000 but if there's a regular pattern, it's very difficult to discern. Thing is...I think these new caps with the 9000 are softer, for some reason...? Anyway, they're all attached. Compressed using x264 at crf-13, which hopefully should be enough. Looking fwd to your replies! |
4 Attachment(s)
The 9800 caps might look sharp, mainly because the contrast range is beyond legal limits for digital video. Makes for a snappy photo with blown out brights and grimy darks, but it's unpleasant viewing. The opening 20thFox logo has ugly line twitter, bad aliasing and shimmer on motion, and stutters badly during the zoom. In most other scenes motion is on the edge of choppy. 2pass VBR encoding will give better detail and motion control than low CRF. Long GOPs like those you have with 250-plus frames will lose detail, will upscale poorly on big displays, smears motion, are not allowed for many playback formats, and not handled well by some playback systems. The encoded bitrates are too low for clean rendering of fast motion video.
There are serious image flaws whose cause and fix can't be determined because no one knows what your original captures look like. All three "progressive" samples look to be processed incorrectly, with serious problems that are kinda obvious and which should have been spotted before encoding. http://www.digitalfaq.com/forum/atta...1&d=1444360812 http://www.digitalfaq.com/forum/atta...1&d=1444360869 http://www.digitalfaq.com/forum/atta...1&d=1444360901 http://www.digitalfaq.com/forum/atta...1&d=1444360947 |
Quote:
Quote:
Quote:
Quote:
Quote:
|
I don't suppose these encodes are good enough to pick out the minute differences between them. So I'll either redo them properly or organize a way for you to get the originals, once you chime in.
|
Quote:
|
1 Attachment(s)
Quote:
Likely that opening Fox logo has a different structure than the rest of the movie. It looks like one of the stock logos for film and TV that somebody screwed up. But it might be possible to smooth things out. The rest of the sample is indeed a trailer from movie source. Appears to be another bad film->PAL->NTSC conversion. There must be a dozen variations in the way those things are processed. I record some HD BBC shows that use plain 3:2 or 2:2:3 pulldown that's easier to break down, but I still run into plenty of oddball structures with PAL->NTSC (HD broadcasting ain't always what it's supposed to be). You'd need only a few seconds of YUY2 unprocessed AVI for analysis, so you could make one sample with that stupid logo to see what it looks like, then a separate short sample with "regular" video movement. Either of those would fit the 99MB upload limit and give us better info to work with. As for long GOP's: these are normally employed for very small file sizes for internet streaming. The priority is file size, not quality. One GOP in your samples was encoded at 277 frames. For comparison, remember that DVD and BluRay spec were designed for assured playback and good motion rendering on both the best and worst playback systems. The accepted maximum GOP for DVD is 18 frames. For BluRay, the spec requires either 1-second or 2-second GOP's (that's 30 or 60 frames NTSC, 25 or 50 frames PAL, or 24 or 48 frames for film speed). Color looks cleaner and smoother in the BJC cap and it looks like either less edge noise or less phoney sharpening. These are subtle differences, but color in the other two caps looks a bit more muddy (it's starting to look grungy in empire_1_9800). If you have your AG-1980 sharpness turned all the way down I'd say it's not necessary to turn it all the way off. Otherwise there are better sharpeners in Avisynth and VDub. I've been working on a cap from a 1980's VHS of Fantasia. Yeah, it's out on newer DVD/BluRay issues but they're out of print and selling at robber baron prices or screwed up pirate issues and, frankly, having seen one I noticed digital problems of their own. You'd think a VHS of a 1940 movie would be simple enough to remove telecine and fix up the noise and bad color. Telecine was no problem, but in several scenes the borders change size (???) and in some shots the animated critters do some odd stuff like one frame is 2 frames ahead of the others, in another spot a frame is repeated later. This happens on all three of my VCRs and on TV play, so it's definitely in the tape and not a problem during capture. It's just a sloppy film transfer, period. I could live with the digital version, but frankly this is a self-learning exercise with Avisynth. This stuff doesn't happen just on tapes. I have a DVD recording of a Vitaphone 1928 short from TCM with frozen and missing frames (I saw it happen during the broadcast. What a mess! And telecined, too). If you think you have some frame structure problems, yours are rather common. Try working with this unprocessed s-video capture off SD cable: BlossomSeeley_and_BennieFields_1928.mpg (14 secs). Apart from the logo in your sample, the movie parts likely have consistent processing. But it's hard to say without a chunk or two of the original cap. |
6 Attachment(s)
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
|
Quote:
|
Quote:
Most people run Avisynth scripts that cut up a video in different ways (double rate deinterlace with QTGMC or bob() or yadif , ivtc, sRestore, SeparateFields, unblenders, pull tricks like changing the field order, etc.) and watch the frame by frame results in Virtualdub. There's an old Neuron2 guide posted earlier for analyzing video having many odd structures -- not everything, but it's a start: http://www.digitalfaq.com/forum/atta...-analyze-video. Sometimes trying to fix a piece of thoroughly ruined video isn't worth it. Often it can't be fixed, or it'll just look worse. The Fox logo is plain 29.97 interlaced, Well, it was formerly interlaced or telecined, then deinterlaced or un-telecined incorrectly resulting in edge shimmer and aliasing on movement, then reinterlaced. The rest of the preview or trailer was originally film speed. At one point it was speeded-up PAL, at another point NTSC telecine or PAL dupes and pulldown added to get NTSC, at another point field-blend deinterlaced (sorry, it can't be repaired), with apparently some shots resized while still not made progressive, etc., etc. In other words, whoever put that trailer together was creatively destructive. Maybe the actual movie content is built more carefully, but the trailer is best left alone and encoded as 29.97 interlaced even if some sections are blended-field "progressive". It's done often with movies that will never again look like the original. You can use ivtc or QTGMC/sRestore to remove some duped (pulldown) fields but the field blending will still be there, and the Fox logo couldn't be joined to it because the logo is a different structure and frame rate. The logo isn't worth deinterlacing because the field data that formerly described smooth edges during motion are gone forever. The logo can't be ivtc'd with the rest of the trailer because the logo isn't telecined. So you could keep the trailer and logo and correct color or whatnot if you want, but it's interlaced 29.97fps all the way if you want it all in one piece. |
Quote:
Quote:
Quote:
Quote:
Quote:
Anyway, thanks for all your comments - they're educational! Do you have any more comments about the difference between the three caps having seen the Huffyuv versions? |
Quote:
|
1 Attachment(s)
Quote:
Field blend deinterlacing usually deinterlaces an interlaced frame into two fields, expands the fields to full-size images, then combines the images into one frame. Each pixel in the blended frame contains data from both original fields. They can't be separated. Quote:
Quote:
Shots like that Fox logo were (and still are) often created as interlaced video. Some are simply progressive but encoded as interlaced. Other animations are created at something like 15fps, and then a complex operation is performed with both duplicate frames and telecine applied to attain a certain frame rate. Some logos are created as 23.976 progressive, then pulldown or frame dupes are applied to get 25fps or 29.97. There are many variations. Many of them just look screwed up. When you add film-to-NTSC-to-PAL and other variations, you get real oddities. Sometimes you can fix it, sometimes not. They're trailers and logos that many labs just consider too inconsequential for proper work. The Foxlogo runs at 29.97fps whether interlaced or single-frame denterlaced with yadif. You can ivtc the other shots at get 23.976 out of it with some blended frames remaining. But you can't combine 29.97 and 23.976 into the same clip (Ah, but see next paragraph below). So the entire trailer including the logo were configured to run at 29.97fps, start to finish. Many trailers are made this way from disparate video segments. You see it all the time in TCM promos shown between main features. It's possible to partially ivtc the video part to run at 23.976 (with some blends remaining), and to break some rules with the Fox logo and make it run at 23.976 progressive. One could use one of the deinterlacing and blending anti-alias filters that works in a fashion similar to yadif mode=0. This does knock out some fine detail but smooths aliasing (insofar as possible with that crummy logo) and outputs same-rate video. The script also repairs some of the bad splits ("ripples") in a few frames and replaces the last bad broken Fox image with the previous good one. BTW, it doesn't take a keen eye to see the obvious horizontal rips in the original: Code:
AviSource("Drive:\path\to\empire1_fox.avi") Quote:
Quote:
|
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
|
Quote:
When used on interlaced video, santiag uses NNEDI3 to produce an effectively progressive video using the same method used by yadif mode=0 but with fancier interpolation to create new frames. It's also sometimes used on video that's deinterlaced first (with somewhat less effective results). There are other anti-alias filters that can use simple low-pass methods (downsize/upsize/blur) or more complicated operations, but they're more destructive. QTGMC was used to smooth object flutter and line twitter during motion. In default "slow" mode it removes fine detail in "real" video. |
Very cool stuff. It's been educational, as always :)
What's your opinion of santiag as a general AAer for progressive video (not deinterlaced - 'natively' progressive)? Or as an AAer/deinterlacer for interlaced animation? |
Site design, images and content © 2002-2024 The Digital FAQ, www.digitalFAQ.com
Forum Software by vBulletin · Copyright © 2024 Jelsoft Enterprises Ltd.