Whats convultion3d like for dvd source?

[BluDChyLD]

Hi,
I've read about the new c3d filter, it seems very good for tv and vhs capture.

What about dvd source? I know it's supposed to increase compressibility, has anyone tested how much?

[JohnMK]

I did a little testing. 1-pass, quality 100%. The Sound Of Music DVD. Recommended C3D settings for DVD movie (0, 8, 8 . . etc., I believe) resulted in a 3% drop in file size, from 101MB to 98MB.

Somebody tell me if I'm using the right compressibility test technique?

[iago]

@JohnMK

Recommended C3D settings for normal/clean/HiQ DVD is not (0,8,8,8,8 ... etc.). Please take a look at the convolution3d.txt file in the latest v1.01 version.

Also, what method are you using to measure compressibility? (I recall we'd already discussed it in another thread before ;))

iago

[JohnMK]

100% quality, 1-pass, file-size comparison.

[vlad59]

Wouldn't it be better to use a constant quantizer 2 for your test ????

[Emp3r0r]

aren't those the same?

[vlad59]

I remember reading a post about that, I'm not totally sure that it's exactly the same.

Can't remeber where and when I read that ????? :confused: :confused:

[JohnMK]

100% qual == const quant 2.

[neo_sapien]

convolution3d 1.00 will increase your compressibility quite a lot.

In the case of Convolution3D (0, 3, 4, 3, 4, 2.8, 0), on the movie "Joe Somebody" @ 640x352 1CD rip and DivX 5.02 with b-frames, the compressibility was increased from 42% with no filtering to 77%.

In the case of Convolution3D (0, 6, 10, 6, 8, 2.8, 0), on the movie "A Bridge too Far" @ 640x272 2CD rip and DivX 5.02 with b-frames, the compressibility was increased from 62% to 120%, and I was able to do the movie at 1 pass quant 2 and still undershoot my target 2 CD size by about 120 megs. I could have done the movie on 1 CD at 450kbps/sec at that same resolution at 52% compressibility test, but that wouldn't be right on a 176min movie.

The movies look great. They're not as smooth as you'd imagine they'd be for roughly twice as much compressibility. Vlad59 should be given a friggin' nobel peace prize for introducing this wonderful filter to the world and giving us DVD presets.

If I could game out a bad quality TV capture preset that produces no ghosting artifacts, I'd be happy as a clam in a bordello with this filter.

[BluDChyLD]

thats a heck of an increase! Were the movies noisy or clean?

[Boulder]

I've used Convolution3d with TV captures with the settings (0,10,24,10,24,3,0). As you know, it depends a lot on the quality of the source, but I got rid of most of the analog noise and still kept the details. Settings (0,10,16,10,16,3,0) would probably be enough but I increased the chroma threshold to increase compressibility. I noticed that luma thresholds below 8 were not enough in my case, and 10 removes noise as much as possible without removing all the details. I resize with LanczosResize to get some sharpness back.

[neo_sapien]

@BluDChyLD

On a scale of 1 to 10, 1 being the noisiness of "Star Wars: Attack of the Clones" or another purely digital movie and 10 being "The Seven Samurai", AB2FAR was a 6 or 7 and Joe Somebody was a 2, being a new DVD from a movie that was released in theaters in 2002.

@Boulder
Thanks for your suggestion :) I'll try it and others wednesday night when Enterprise comes on. It's a local broadcast station on a cable channel that is analog, so there's lots of noise.

[BluDChyLD]

that's impressive, I thought maybe you were using really noisy sources to get that kind of compression increase.

I'll try it myself and see what results I get, if it does smooth the video too much I'll use it more often :)
BluDChyLD

[JohnMK]

How do you determine your compressibility figures, neo_sapien?

[JohnMK]

Just anecodtal evidence. For an old movie (1930's Sir Laurence Olivier doing Shakespeare), using preset movieLQ reduced file size by 17% (100% quality, 1-pass). That's, I believe, correct me if I'm wrong, a 20% increase in compressibility (6 divided by 5, anyone?). I encoded the movie @ 55% compressibility, according to Gordian Knot, so I adjusted it in my mind, realizing it's probably something closer to 70-75%, for a truly, great looking movie. :D

[JohnMK]

By the way, I've found lately that 70% quality is a better target value to aim for in Gordian Knot. This looks far more, subjectively better to my eyes than 55-60% with a higher resolution.

[neo_sapien]

Originally posted by JohnMK
How do you determine your compressibility figures, neo_sapien?

Jonny's Comptest 1.02 is an invaluable tool for doing comp tests. Just load an AVS, set bframes to on and psycho v's to whatever level you want (although this doesn't affect compressibility), and press start. Load the resulting log file in Gordian Knot, and voila. Good luck JohnMK :)

[JohnMK]

Thanks! Where do I get it?

[JohnMK]

Here:

http://jonny.leffe.dnsalias.com/

[BluDChyLD]

Originally posted by JohnMK
By the way, I've found lately that 70% quality is a better target value to aim for in Gordian Knot. This looks far more, subjectively better to my eyes than 55-60% with a higher resolution.

I agree, I used to aim for 60 - 65 but now Im using my tv out which is hides the pixellation I use a lower res for a higher compression rate. There's definetley a differance between 60% and 70% but some people prefer to keep detail instead.

[theReal]

I think the main benefit of C3D is that it can improve the looks of copies from mediocre DVD sources very much.

As an example: Sixth Sense, PAL, which shows annoying artefacts even when the compress test is 67% (it even shows these artefacts with quant 2 because they are present on the DVD source and are amplified through resizing and re-encoding).

I added C3D with low settings and increased the sharpness of bicubicresize a bit (from 0,0.5 to 0,0.6). The compress test went to 70% then, not too much of a gain, but the result looks so much better because all the annoying artefacts from the DVD are gone.

[jarthel]

Used convolution3d on anime. Gives good results, faster than SShiQ. only complain is smoother strength doesn't have much option. :)

jayel

[sapient]

I don't want to sound negative, but my results with convolution3d have been rather disappointing.
For the PAL version of the Time Machine (2002) convolution3d increased compressibility from 48.6 to 50.4 using Lanczos or from 60.0 to 61.9 using bilinear. I have had similar results with some episodes of Friends on DVD. The only time convolution3d gave me worthwile results was with a really old tv show transfered recently on DVD, presumably from a bad source. Given the fact that convolution3d increases encoding time by 50%, a less than 2% increase in compressibility for a clean DVD source makes it really not worth my time. So, as far as I am concerned, convolution3d is a really nice denoising filter, useful only when you have a... noisy source.
Nothing more.

sapient

[OvERaCiD23]

You have to tweak C3D quite a bit for each film, but the results can be amazing. Once you have the settings perfect (little blur, no ghosting), it can drastically improve the quality of your rips. I don't see how one could not like this filter (one of the best IMO), but everyone is entitled to their opinion.

[JohnMK]

If people want a little bit of extra filtering on a clean source wouldn't it be better to use bilinear resize instead of c3d? Bilinear resize is significantly faster, and accomplishes a similar effect, no?

[theReal]

Using Bilinear Resize results in washed-out colors and blurriness that can be avoided by using C3D.

I'm still convinced that C3D's main benefit is not the increase of compressibility, but the better quality of the encoded movie. I've done four movies that looked significantly better with C3D than without any filters. That is regardless of size - I compared results with quant 2 and quant 4 before I decided to use C3D because of the better looks and not because of better compressibility.
An increase in compressibility was also noticeable, but not by too much.

Of course this won't work on perfect DVDs - they just look best without any filters. However 90% of DVDs are far from perfect and can benefit from a little C3D...

[MaTTeR]

I strongly agree with The Real, most of the DVD's I've seen are garbage quality except for those fabulous SuperBit transfers. Bilinear gives much too soft of an effect for my eyes. I recently ripped "Last House on the Left" which I'd vote one of the worst DVD transfers seen, however when I applied C3D the rip came out looking better than the original DVD. The filter does some amazing work depending on your source and settings.

[BluDChyLD]

Originally posted by MaTTeR
I strongly agree with The Real, most of the DVD's I've seen are garbage quality except for those fabulous SuperBit transfers. Bilinear gives much too soft of an effect for my eyes. I recently ripped "Last House on the Left" which I'd vote one of the worst DVD transfers seen, however when I applied C3D the rip came out looking better than the original DVD. The filter does some amazing work depending on your source and settings.

I'll have to try this filter out on some of my cheaper dvds. I wouldn't of thought a film like "Last House on the Left" wouldn't get much of a good transfer, being a video nasty ;) What are the best setting to use with c3d for a "normal" film?

[sapient]

I think that whether c3d improves the quality of a movie or not is more a matter of taste than a case of true improvement of the source. Personally, I do not like the somewhat hazy, smoothed effect that c3d produces and so I avoid it when there is no significant noise in the source