Divx script vs Optimal script?
I know the reasons for the 2 different versions but have a few questions.
If you have a pretty clean divx is it better to use the optimal or just use the divx with a few variations. I did a test with my latest divx and this is the results divx V3(as is)-Encoding... CQ : 56.837-min 634 max 2500-High Quality divx V4(as is)-Encoding... CQ : 53.787-min 634 max 2500-High Quality Optimal script(as is)-Encoding... CQ : 70.000-min 634 max 2000-motion estimate THese were with Tok and i did a full encode on the OPtimal script(figuring it would be wrong)and it was almost right on target THats quite a difference in cq between the 3 scripts. All suggestions welcome Thanx |
I never had V4 giving lower CQ than V3, nor optimal giving such HUGE CQ (comparing to others two). But the diff in MAX bitrate used is HUGE itself !
Do the same with the same MAX-MIN settings for the third one than for the two others. BTW, wht did you call "optimal" ? The MA one ? I hate its bluring effect and I do not use it even on a DVD that is clean. So on a yet blurry divx... never. That's only my choice. |
Also did some tests of my own:
Scripts used: DivX V4 (Latest) DivX Deen (used: deen("a2d",2,8,10) instead of Convolution3D(1, 6, 12, 6, 8, 2.8, 0)) MA (Latest) Clip was only 2 min. (min bitrate=481 is 0.57*avg bitrate for a full encode) All encoded with a CQ of 60. 1. DivX V4 min=300 max=2300 - Size= 8.995 KB 2. DivX V4 min=481 max=2000 - Size=10.517 KB 3. DivX Deen min=300 max=2300 - Size= 8.864 KB 4. DivX Deen min=481 max=2000 - Size=10.396 KB 5. MA min=300 max=2300 - Size=7.883 KB 6. MA min=481 max=2000 - Size=9.659 KB Visual the DivX V4 and DivX Deen is about the same and the MA is a little more blurry. So this would lead me to say that using Deen would give a little higher CQ using the DivX script, but this also shows that using the MA script will even give a higher CQ (but a more blurry picture). I would agree with Dialhot that in scenes with a lot of action in the background, the MA script will blur the complete scene too much, because even if there is a lot af action in the scene you will focus on the front, where there might not be so much action. Another thing I've been thinking about is, has anybody tested if a low CQ with min=0.57*avg - max=2000 is worse or better looking than a higher CQ with min=300 - max=2300?? More tests to come... Anerboda |
I have an avi that gave me a file 10 MB too long (145 insteed of 135 wanted). I will try tomorrow the exact same file and Tmpgenc settings but with "deen" insteed of "Convolution3d". And I'll tell you if my target size is reached and which I find visually the best one.
Note: When you do some tests, use a min CQ of 70. 60 is considered as too low by a lot of people, including me. And the image quality is to poor that it is difficult to evaluate tiny things as details losses in the hair (the most difficult thing to render - and where I focus when I try new filters) |
Quote:
As I said in the spanish section too ... It depends on the movie contents .. cause by lowering the max, the bitrate has less space to "breath" .. and in case of very high action scenes or when the image contains a lot of parts which rapidely change between light and dark (Sunlightreflections on moving Watersurfaces or diamond reflections for example) the image in these parts of the movie could end with blocks! So you have to assume what to to ... when encoding a calm movie or a movie which could contain high bitrate peaks. And also it depends on which Tv device you use ... high quality Flatscreen will show much more worse elements within the picture as a few years old standard Tv. Also ... if you recognise that by lowering the maxbitrate your CQ rises up .. that means also that at some high VBR parts of the movie the "peaks" will be cut or kept down! But as I figured out, sometimes there are just little parts in scenes which you even won't recognise but on the other side they perform for just a a few frames a "jump" in the VBR scala. Well and thats what I meant .. you have to try .. cause even with these a bit cuttet very short time taken "jumps" and therefore using a higher CQ you still can receive a very good looking movie even when watching these mentioned scenes! Cause when using higher CQ the video in average will look better .... but that's nothing new I think. :wink: |
Okay. Very quick tests on that file I talk about in my previous post. With deen the file size dropped by 4 MB and a quick glance at it doesn't reveal any difference in visual quality.
For sure I have to make other tests before decinding if deen will be aprt of script V5. |
More tests:
CQ 70 Size 544x576 Length of sample: 1 min Max bitrate=2300 , min bitrate=300 To judge the visual quality, i opened the samples in Vdub, and compared the same frame, looking for the details in a beautyful blondes hair... 10 being highest Scripts used: DivX V4......size=10.966......encoding time 2:30 min......Visual Quality 10 Deen V1......size=10.804......encoding time 2:29 min......Visual Quality 9 Deen V2......size=10.780......encoding time 2:47 min......Visual Quality 8 Deen V3......size=10.897......encoding time 2:29 min......Visual Quality 10 Deen V4......size=10.874......encoding time 2.47 min......Visual Quality 9 MA..............size=10.643......encoding time 2.55 min......Visual Quality 8 Deen: (instead of Convolution3D) V1 = deen("a2d",2,8,10) V2 = deen("a3d",2,8,10) V3 = deen("a2d",2,6,8) V4 = deen("a3d",2,6,8) Conclusion: Slightly higher CQ with deen("a2d",2,6,8) almost same encoding time and same visual quality as DivX V4(to my eyes anyway). Doing a full encode (movie is 82 min): Prediction with CQMatic gave a CQ of 70,08 Wanted filesize=749.928 KB Encoded filesize=751.080 KB diff 0,1536% Not bad !! But then again, why use Deen instead of Convolution3D, when there is nothing to gain? To me there is almost no difference. The only thing I've found out by these tests is that filesize prediction went very well... But I will have to test it on other sources offcourse. I just wanted to share my testings and I must agree with Dialhot that his script for DivX files is still producing very good results, I'm sure he will bring it even further... Anerboda |
Thanks for all you tests Anerboda :-)
|
Thanx for doing the tests Anerboda
@ Dialhot Sorry for taking so long but i have been a little busy and haven't had the time to do the test with the same min max until tonight,and the optimal script i was refering too is the MA script. Here are the new results divx V3(as is)-Encoding... CQ : 56.837-min 634 max 2500-High Quality divx V4(as is)-Encoding... CQ : 53.787-min 634 max 2500-High Quality MA script(as is)-Encoding... CQ : 63.994-min 634 max 2500-motion estimate I only did the min and max in the start of this thread that way because i thought thats the way i was suppossed to. What are the recommended min and max for the divx scripts |
Quote:
What is the movie you encoded ? (PM me if you want). I have to test this one ! Quote:
|
Hey Dialhot
I'm going to try the test again with all the same features,i was using what was recommended for each script(earlier version of tok with detect scene change with MA script and the newer version of tok with the avi script) One other thing when i use the MA script i have to use BicubicResize(496, 334, 1/3, 1/3, 9, 0, 590, 336) instead of Gripcrop because i get errors but if i try to use BicubicResize in the V3 and V4 scripts i get an error Code:
Sorry DCT filter test version needs size multiples of 16x16 Thanx |
Ok with all the same settings except the gripcrop problem here are results
min 64 max 2300 V3 -Final CQ: 70.769 V4-Final CQ: 70.303 MA-Final CQ: 71.774 A lot closer now |
Okay, so you obtain the same results I have : MA a little above the 2 others.
Now the choice is in the visual result obtained. For my part (my eyes), V4 wins. I can't stand with MA script even on DVD and I do not recommand to use it on Divx that are generally a little blurry. But trust your eyes and take the one your prefer. |
Site design, images and content © 2002-2024 The Digital FAQ, www.digitalFAQ.com
Forum Software by vBulletin · Copyright © 2024 Jelsoft Enterprises Ltd.