I'm converting HD movie (AVC codec, mkv) to XviD and downsizing resolution to 720x528 (keeping aspect ratio close to original 1,37:1). The original is b/w. By the way, what color code would you use? Is RGB24 (888) fine?
Using VirtualDub I went through analysis pass (1st pass) and am now going through 2nd pass. Movie is 2h 10min long and I have estimated 2,4GB (video + 224 Kbps AC3).
The overall quality is great, but in particular scenes I get the following result ...
I would say the analysis pass estimated bitrate that is too low (the picture is still). I have tried adjusting the Curve compression High -5% & Low +10% (bellow are the default values). It didn't help.
I have tried specifying zones, increasing weight and no use.
What would you do? Increase the overall size? Hoping for some guidance
+ Reply to Thread
Results 1 to 12 of 12
Thread: Blurry Frames
Last edited by manono; 13th Nov 2012 at 15:17.
Thanks for the reply and the advice!
It is close to black. Different dark shades of grey and also grainy. I would like to preserve these shades, but instead they get blurry (or however you call it). The picture was downsized, maybe you can see better here: http://imageshack.us/a/img26/8919/converts.jpg
The thing is, I am satisfied with overall quality but would like to improve still dark scenes like this one above.
The weakness in most high compression codecs is in dark noisy areas like that. Use more bitrate in those areas. or filter out some of the noise.
Exactly, how would you increase bitrate in just scenes like that? I tried zones, but it didn't help. Maybe I should use different (higher) settings.
Like I mentioned before, I'm doing 2 pass encoding, and this one is pretty much optimized to give more bitrate where needed ...
Apparently, 1.5x weighting isn't enough. Go higher.
Keep in mind that a weighting of 1.5 doesn't mean that section will get 1.5 the average bitrate. It means it will get 1.5 times what Xvid would have otherwise decided to give it. Since it's a very dark shot it would normally be allocated a very low bitrate. You can verify this after encoding with Bitrate Viewer.
Last edited by jagabo; 13th Nov 2012 at 16:33.
Denoise before resizing. Do it in YUV. GradFun2DBmod is often used to smooth this sort of thing, or even DeBlock, but many say that Dither() is the way to go for these tough blocking problems. You can always use VDub's coring filter at RGB 16, but noise above that value will still be there. This gunk is difficult to clean, even with high bitrates. The problem isn't really banding, even though the link I'm giving has banding in the tile, but it's about getting better flat areas...so take a look at this thread: http://forum.doom9.org/showthread.php?t=153589 .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
Thank you sanlyn!
I'll take some time and figure out what you recommended. It's pretty new to me (that dithering, deblocking, banding etc.) ...
Me too. I first saw those images in a bright room. At first I was just kinda ? ? ? what's the problem, anyway? Didn't look again until tonite. And then I said, yep: same thing that always drives me nuts with big flat areas. They look nice 'n smooth in AVI. but then the encoder hits 'em. Frustrating. Still looking for better answers to that one, but I haven't learned to work with dither() yet. I read where it's a likely solution. Meanwhile I throw deblock at it and add some fine grain. But there has to be a better way.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 made a sample with a dark grainy section (similar to that pictured by the OP) in the middle. With no particular optimization the bitrate of that section turned out to be about 1/4 the average of the entire video. I thought I could just force a high weighting on that section to get more bitrate allocated to it. I found that Xvid's weighting doesn't change much beyond 2.0. At 2.0 the bitrate of that section increased by about 50 percent but that wasn't enough to retain the grain. I did find that you can force a quantizer rather than a weighting. Forcing the quantizer to 1.0 caused it to retain pretty much all the grain. With 2.0 the posterization artifacts were back, about the same as using a weight of 2.0. I found a forced quantizer of about 1.3 was a fair compromise.
Of course, if your video has lots of sections like this it's probably too much trouble to define a zone for every one. Switch to x264 encoding and use the aq-strength setting. That does something similar but it's automated.
Last edited by jagabo; 13th Nov 2012 at 19:22.