sanlyn, you're incredibly persevering
I'm so grateful for everything you did on this video!!
While you're trying new tweaks, I'm testing each step of your script. The thing that makes me curious is RemoveDirt().. I compared carefully but I can't see its contribution Please check the following frames:
+ Reply to Thread
Results 91 to 120 of 124
Thread: remove noise and block.
This picture is so bad that I would record it to VHS and than filter it in hardware. You can also make it in B&W, that sometimes helps on horrible videos.
The 640x480 is another sign for me on a crap video.....Maybe it is just me.....
You can more than likely find a better quality recording of this video somewhere else.
Yea I would be interested in finding out how to fix these motions problems. Downloaded a few FLV files from the NET of super rare recordings.
Was going to restore them, because they are rare. Did a few of them with ok results.
TWO recordings had major motion problems.
As you know FLV files pretty much suck.
A few of them have random moving digital pixels (which is really bad compression), one of them is so bad, that I can't fix it.
Because the quality of these videos suck so much, pretty much do this backwards and re-record many of the videos using hardware and hardware filters to fix some of the problems.
The audio for the most part is terrible. Actually created an audio adjustment for these junk FLV files.
Back to the motion problem:
For the life of me, could never figure it out. It is kind of weird, cause on the PC, you can watch them ok, but on the TV, it is point blank horrible. Tried everything from progressive video to changing the frame rates, to merging frames to whatever, nothing worked. The two videos are PAL recordings, it really doesn't matter, cause who ever created the online video had no clue what they were doing and wrecked the motion on the video.
ALRIGHT I WILL HAVE A GO AT THIS VIDEO, I AM DOING IT MY WAY AND HOW I SEE FIT IT WILL BE IN B&W, AND I AM NOT RESTORING THE SOUND ON IT. HOPEFULLY IT WILL BE UNDER 100 MEGS SO I CAN UPLOAD THE FULL VIDEO.
EDIT: IT IS A 5 MINUTE VIDEO NO WAY IT WILL BE UNDER 100 MEGS, NO IDEA HOW TO DO THIS, CAN SPLICE THE VIDEO UP & YOU COULD PUT IT BACK TOGETHER....GRANTED YOU HAVE TO RE-CODE IT BUT WHATEVER.....OR I CAN UPLOAD IT AS A WINDOWS MEDIA FILE.....
Last edited by Deter; 21st Mar 2012 at 23:56.
Ok whatever can't upload the NEW video cause the file is too big, had to create a YOUTUBE account. On a side note kind of ticked at YOUTUBE couldn't create a custom name for the channel. Was going to do like a Video Help YOUTUBE channel. Anyway, I recorded the video to VHS and decided to work from that. To me it got rid of many of the un-natural artifacts in the video. The quality of this video is never going to be good cause the source was horrible. This was what I was able to do with the video, not going to waste any more time on it. Also cause I thought the colour was terrible converted it to a B&W video.
Not really sure what you guys meant about motion problems. In the video I saw one of the sections was edited with a crossfade which creates a weird motion and a in a few other sections it looks to have dropped frames. That was my take on it.
Please note cause we are using YOUTUBE compression on an already bad video, please watch this in 720p. The reason why it was upconverted was to get slightly better compression.
Last edited by Deter; 22nd Mar 2012 at 11:05.
RemoveDirt is being used to address noise issues, not grain or sharpening. It also does a little work on chroma noise, which you can see on the music stands (red blotches on black). In the first image below ("With_RD") what RemoveDirt is doing isn't very apparent from still images, which can't be used effectively to analyze what motion-block processing is doing. I'm looking at various mode switches in RD to see what will happen on motion, but it is doing some cleanup as-is. These images are PNG's made from the latest effort before NeatVideo (it's Step2, folks) and without a couple of small plugins I'm trying from the anime world. The best way to compare very similar images is to mount them in Photoshop or PaintShopPro (or some graphics app that uses layers and will let you turn layers on and off to see the differences), and enlarge them with bucubic resize (do NOT use "Zoom" controls). But you can click on them to enlarge slightly in the forum, then click the back and forward buttons. In the capture that uses RD, the music stands look cleaner and some of the blotches are smoothed.
If anything, RemoveDirt softens a bit (look at the fine mesh in the microphone). That's easy enough to fix. Watch the moving video to see what happens to noise. In the scene that follows this one, look at the music stands, the components against the wall in the background, shadows in MJ's hair and jacket.
With Remove Dirt:
You're right. Not much difference in the stills. No motion.
I just spent a few hours using various forms of QTGMC. By golly, the darn thing just doesn't do what we want. Sharper image, but it creates block noise and motion artifacts where there was none before (I used QTGMC(inputtype=2). After all the trouble I went through to smooth some of that stuff, QTGMC puts half of it back.
Oh, well...still making progress, though.
ED: BTW, I've been looking mostly at color-corrected ftames of this video on my PC. Every time I see some of the 'original' unprocessed colors I keep saying OMG, there's NO WAY that original video could have looked like that.
Last edited by sanlyn; 23rd Mar 2012 at 01:22.Our inventions are wont to be pretty toys, which distract our attention from serious things. They are but improved means to an unimproved end. -- Henry David Thoreau
You asked if I would give this a go. Did you watch the video I posted above? The picture quality on this video is terrible only so much you could do, plus it was recorded on someone's home camera. Worst of all it is an internet file. Didn't use any filters or Remove Dirt, just recorded it to VHS and it cleaned up some of the digital destruction done to this video. The pixelation to the colour is just flat out terrible. One thing I learned from Lord Smurf, you need to capture the video using the best possible hardware, if it is not done somewhat correct the video is destroyed. This took 20 minutes of actual work to do. I deal with tons of videos and you need to decide weather or not it is a good idea to waste time on it. Something like this video is a waste of time, cause it is never going to be good. The video I did above is more than acceptable considering the source quality. Yes I did make it B&W but the colour sucked on this video.
Hiya, deter. You caught me on my dinner break.
? I don't know what that statement means. I could do the same thing in VirtualDub with two or three plugins.
It was shot totally in available light. A pro wouldn't have done it that way. I think we made some serious improvement with the color and levels. But, yes, the idiot photog burned up the highlights. We can save a few of those. Well, I shouldn't use the word idiot. The kid who made it seems to have an eye for interesting shots, and some of them are kinda cool. Needs to learn something about exposure, though. I learned, but that was part of my M.A. in film production.
You mean, like recording badly compressed and damaged m4v DivX/UTC to even lower resolution on a VCR? I have a great and long-time admiration for LS. I doubt he'd recommend that technique. But I agree with him, and with you: the clip is mishandled muck that most people would ignore.
But thanks for taking a shot and sharing your video. Gotta get back to my research in doom9 concerning this video. Have you considered getting a download account at an online site? I'm in my 3rd year at 4shared, the service is good and it's not expensive.
Last edited by sanlyn; 22nd Mar 2012 at 20:27.Our inventions are wont to be pretty toys, which distract our attention from serious things. They are but improved means to an unimproved end. -- Henry David Thoreau
Yes Lord Smurf has said in his forums he has used that method before. You don't do it all the time. A few years ago, got in a rare recording from a crapy VCR on a crapy DVD recorder. It was also recorded in 4 hour mode. The picture was a nightmare and had a lot of "digital destruction". The Macro Blocking was really bad also. Was like what the hell am I going to do with this.
Found using some of my hardware as a passthrough was able to alter or fix the picture a tad. Now having some pretty nice VCR's decided to try re-recording the video back to VHS. Played back the DVD and sent the picture through a Time Based Corrector. Recorded in SVHS mode in SP on a JVC deck.
It was able to clean up the video enough and not have it look like something fake. However it tends to dull some of the detail a bit. Sometimes with horrible videos bluring the problems helps.
Because VHS is analog it got rid of all the MACRO BLOCKING. Than took the video again and played it back on a high end VCR and ran the picture through filters. Was able to rid myself of a lot of the problems with the video. This video turned out really well.
Started messing around with this method to restore some FLV files. However was able to get better results just using hardware and re-cording it again in digital. However it depends on the recording. In some cases the VHS re-record looked a lot better.
This Michael Jackson video is terribe, so the results are going to be blah...Last night when I was testing this out it looked better on VHS.
As far as re-colouring video, last year did like 30 recordings from the 1960ties it was a nightmare. The videos had way too much red....
Have an FTP server that I use but not posting it in a public forum......
I should have an improved version by, say, Saturday eve or Sunday. Not a quantum leap, but improved. And a few tweaks of the script.Our inventions are wont to be pretty toys, which distract our attention from serious things. They are but improved means to an unimproved end. -- Henry David Thoreau
I hate to say it, gang, but we had a power failure in my neighborhood for 7 hours due to high winds out in the county. That's one of the hazards of living near a small body of water called the Atlantic Ocean. No power until an hour ago. I went to a friend's house to use their laptop and get on the 'net, but fell seriously behind on this project. Oh, well...will be at it tomorrow.
Running off examples some samples now. Saints, what an experience ! I've spent so much time with QTGMC (supposed to rebuild damaged damaged interlaced and/or progressive video, for what that's worth). I ran QTGMC, the existing filters, so0me new ones, etc., in more configuartions than I can number. Used up 47-GB of disk space doing it.
My experience with QTGMC on this video is that no matter how you set it up, you lose more than you gain. I have to agree with an earlier assessment: this video began life as interlaced DV. Somewhere along the line someone tried to make it progressive, apparently by removing half its frames (!). The more you fiddle with this creature, the more it looks as such. This means that 50% of the original video data has disappeared. Further, it means that the remaining YV12 pixels, rather than 2 to 4 pixels apart, are now 4 to 8 pixels apart. I've even run plugins that use mvtools and set the mvtools search parameters to 4 and 8 pixels apart (the default search is 2 in each direction). The result is a blurry mess. I'd prefer to keep much of the noise; at least you can recognize all the objects.
I learned many things, one of which will prove handy in the future. By default, QTGMC uses NEEDI3 to deinterlace. I've never liked that plugin. The result with this video is that QTGMC relieves a few artifacts but creates many more -- and did nothing to relieve the shimmer it's supposed to address. However, I did find that one can substitute a few other deinterlacers. The best turned out to be yadif. You use yadif with this statement (here, used for progressive video):
I'm sticking with the old filters, but I reversed order on some of them and ran TemporalDegrain first. Better detail than before. And I used SmoothLevels this time. Running scripts on and off all day today, and will post some samples. I can't say it will be so much "better". Just "different".
I'm looking forward to see your new version sanlyn
I agree that QTGMC() might also cause the cartoon unnatural effect.. and anyway you could get rid of the combing artifacts with other filters.. But did you think of using more simple deinterlacers like TDint()? Although we'd still have the loss of information due to interpolation..
So you think they removed a field to make the video progressive? Why would that make the YV12 pixels apart from 4 to 8? Are you talking about the fact that in YV12, a pixel carries the chroma information for its neighbor? But to remove one field shouldn't mess the chroma samples since they are assigned one line over two on interlaced content... Not sure what you meant here though..
I'll check the contribution of removedirt on motion, looking carefully at the details you mentioned. But on still, I agree that there is not much different.. to tell the truth, I have to stick my nose to the screen to see the differences.. so I think such improvement is meaningless.
I had a look at smoothlevels and I wondered why you used the parameters : TVrange = false, Lmode=1, brightSTR=1,DarkSTR=5. Could you elaborate?
I'm also curious to see your settings for the colors on each scene. When they are on the room in front of the mic, the colors really look dull. What you did so far was above my expectations and I'm so glad you didn't drop the video just because it was so bad. I think that it's on poor sources that we can make the best improvements!
That gritty grain must have been added. I can't imagine what would possibly have placed that awful grain in there, other than software designed to do it.
NeatVideo, and decided MergeChromas did less than I wanted. The idea was to improve color depth and reduce chroma noise in shadows, but removing MergeChroma accomplished the same thing in this case. I didn't want "hard" sharpening on luma, but also in this case it made no difference without MergeChroma. I incra4esed RemoveDirt's strength, which sharpened some areas a bit more than I wanted but also removed bigger clumps of simmering grain without affecting detail.
I used settings I found suitable for both the darkest and brightest parts of all the scenes in the video. Some shots could later be modified as circumstances dictate. But once data is lost in the most extreme shots, it's lost forever.
I tweaked the colors a bit. I had to anyway because replacing the former ColorYUV() with SmoothLevels changed some of the colors and levels, especially at the extremes.
I agree: you can't learn much from pristine video that requires little or no effort.
Thank you for your answer.
Earlier in the thread, you said that 16-235 was for DVD standard and out of this range, colors would look saturated and washed out on TV screens...
I just wonder that now because I'm not going to put this file on DVD and it'll only be played on PC. So maybe it would be ok to let the video in a 0-255 range..
It's still not very clear in my mind in which cases I should work between 0-255 or 16-235..
I have to leave my PC soon so I couldn't observe and think deeply about the field removal issue.. I'll look at it later today.
Last edited by mathmax; 29th Mar 2012 at 06:24.
All digital video should be Y=16-235, RGB=0-256.
True. I found SmoothLevels helped keep some of the burned-out highlights under control in several badly-exposed shots. No way to recover completley, auto-exposure and subsequent processing of the original didn't help matters.
After pruning unacceptable versions from my hard drive (there were lots of 'em), I'm down to 11 complete plays of this video. Sat at the PC for 2 hours last nite trying to decide among the least-worst of them. The most viewable seem to be those those with fewer and weaker filters. None of the QTGMC versions are among these.
I'm repairing two PC's for most of today. Will have something posted some time over the weekend.
Last edited by sanlyn; 30th Mar 2012 at 11:54.
This is why my 20 minute work on this project is the answer. You have spent like the last month on this project. Is it really worth it? You can only do so much with a crapy FLV file.
Last edited by Deter; 30th Mar 2012 at 10:32.
There are perfectly good filters to remove macroblocking without resorting to VHS. You could add the same blur as VHS digitally (maybe similar noise too?).
And, of course, there are better quality analogue formats than VHS (pro and semi-pro ones) that would preserve bad macroblocking perfectly well.
But it's an interesting approach. I hope I never have a source so bad that I feel the need to try it though!
AT this point I think it's as viewable as I can get it, without knowing a lot more about very sophisticated masking and motion tracking techniques. Oh, well, that's why I have 93 Avisynth plugins and a new copy of After Effects Pro -- the latter, I better get to work learning how to use. It cost far too much to have it just sit idle on my PC! Will touch up color and stuff tonight and tomorrow. . . .
I just go back to your previous post. Thank you for taking the time to explain in detail.
Midstream was a little better but just a little, and still with new artifacts.
About Smoothlevels, you said TV_range made no difference if it's true or false... but the parameter has a meaning? What about the other parameters: TVrange = false, Lmode=1, brightSTR=1,DarkSTR=5 ?
In which case do we leave a final video to RGB 0-256?
Looking forward to see your new version..
I said TVrange made no difference in that particular case because of the other settings used. The various settings for SmoothLevels are detailed in the doc that come with the plugin (attached below).
As usual on weekends, wife and family have monopolized most of my time (and the power outage last week didn't help). Working on color now and have a few scenes done. I have the house all to myself tomorrow, so I'll make progress 500% faster.
Understood, jagabo. I'm trying to avoid facial burn-out on some of the shots due to in-camera exposure (or bad processing somewhere). Can always modify the color/levels later, but bright red on some of these shots is outta sight.
Working on color, but have a PC client today. Never fear -- be back soon.
After several delays, almost finished with a new post. But . . .
Working on the shots that were made thru several glass panels. Shooting thru glass -- especially when the objects seen thru the glass are lighted much brighter than the lighting in one's shooting position -- you tend to ignore reflections on the panels themselves. Shooting thru glass with high-magnification zooms also involves in-lens astigmatism and more reflections. These problems start showing up when you correct levels and color (zoom lens astigmatism can't be fixed, so detail gets soft. No cure for that, except to pay 3X more for anastigmatic lenses). Those reflections show up as blue or green smudges in some shots. Working on that now. Pain in the neck.
ED: I'm building this version without RemoveDirt. Looking at the older stuff, I now see that RemoveDirt mitigated much of this discoloration. Will just have to live with it.
Ed: I've been staring at this video so much for the past week, my eyes and brain are conditioned to see only what they want to see. Better take up the old colorist's caution to leave it be for a night and let my eyes/brain get back to reality. Will veg-out for a while, watch some Turner Classic Movies. It will all look different tomorrow.
Last edited by sanlyn; 5th Apr 2012 at 17:44.
Okay, it's been a long haul . Many interruptions, including the Easter bunny. The result isn't a quantum improvement, but it's...well, an improvement but not of quantum class. At least most of the horrible grainy gunk is gone.
version 8 preview (NTSC MPEG2 AC3, 127-MB, 2-min 44-sec) 23.976 fps progressive - Not DVD compliant.
The Whole Thing (NTSC MPEG2 AC3, 262-MB, 5-min 38-sec) 23.976 fps progressive - Not DVD compliant.
The download site reports 2 minutes and 6 minutes for these.
The scripts are in 3 steps. I used 3 scripts to avoid freezing-up my pitiful 2.4GHz PC. Steps 1 & 2 are saved as YV12. Step 3 is simply a conversion to RGB24 for VirtualDub and NeatVideo. NeatVideo used mostly as a temporal filter. RemoveDirt() not used, but I think I should have kept it anyway for some green chroma noise. I ran NeatVideo (Step 3) by itself. There's really a "Step 4" for the color filters.
The video has 20 scenes, each color-fixed individually in VirtualDub. How you break up the AVI into 20 scenes is up to you. I used Avisynth and Trim() + Lagarith. Each scene was encoded to MPEG with TMPGenc Plus 2.5, which allows me to cut each AVI so that the first frame in each scene is a key frame, with some 30 frames of the foloowing scene retained for audio sync with the next shot. The scenes were assembled into one continuous video with TMPGenc MPEG Editor v3. How you want to arrange this edit/join is up to you, but TMPGenc made it very easy assemble the 20 MPEG's.
Scenes 3 and 4 were joined with a dissolve in Avisynth. If you start counting frames from frame 0 in scene 3, the dissolve begins at frame 391. My version of Scenes 3 and 4 had extra frames, but basically the trick is to trim Scene 3 from frame 381 + 50 extra frames. Trim Scene 4 from the frame that starts the dissolve. The dissolve is 50 frames (about 2 seconds). Here is the script I used. You'll have to fiddle with the frame numbers. I used an Excel spreadsheet to figure it out. In a folder called sc3_4 I joined scene 3 and scene 4 = "sc3_4":
sc3c=AviSource("J:\forum\AllIn\post8\sc3_4\sc3.avi").Trim(0,464) sc4=AviSource("J:\forum\AllIn\post8\sc3_4\sc4.avi").Trim(28,0) Sc3_4=Dissolve(sc3,sc4,50) Return sc3_4
Later I'll make a 3:2 pulldown 29.97fps version -- Which means I'll have to join all these AVi's in Avisynth to avoid interrupting the 3:2 pulldown sequence. Gimme a little time on that.
Still not satisfied. Check back in a few weeks after I learn more about mvtools and masktools.
Last edited by sanlyn; 9th Apr 2012 at 19:31. Reason: Trim statement modified
Trim(0,-465) is the same as Trim(0,464).
Perhaps you're thinking of the need to use Trim(0,-1) instead of Trim(0,0) to get the first frame, since 0 has a special meaning for the second number.
Yes. Old script style I've used for some time. You're correct, either statement does the same thing concerning frame 0. Old habits are hard to break. I'll update.