TMPGEnc: Motion Search Precision worth the wait?

[Will Hay]

I'm encoding a 62 minute Virtualdub capture with TMPGEnc using the following settings:
DVD PAL (MPEG-2 720x576 25fps VBR 7336kbps, Layer-2 48000Hz 384kbps)
...and I've changed the motion precision tab from motion estimate search (fast) to highest quality (very slow), just to ensure the highest quality I can get.
Trouble is, it's been encoding for almost ten hours and the source position is only 21 minutes!
I know TMPGEnc isn't the fastest, I'm not complaining about that (it's my encoder of choice) but is it worth the huge increase in encoding time for this change in motion search precision?
At this rate the encoding will take 25-30 hours!
I wouldn't say my computer is top of the range but it's relatively a decent spec, and encoding I've done recently at around 3000kbps on the fastest motion precision setting takes twice as long as the running time, ie. 2 hours encoding per one hour of capture.
Is it worth the extra encoding time?
Thanks for any assistance
Will

[disturbed1]

Highest Quality (Very Slow) doesn't offer much improvement over High Quality (Slow) just doubles the encode time.

High Quality will yield better results than the others, with a good speed:quality ratio, much better than the speed:quality you get from Highest Quality setting.

Just set it to High Quality (Slow) and forget about it.

[Will Hay]

Many thanks
W.

[txpharoah]

I suggest ESTIMATE. Unless you're working with video that has lots of panning or is footage of horse races or from the speedway, even HIGH can be a complete waste.

This is my setting of choice, and I do lots of freelance projects from home with this setting. No customer complaints to date. No complaints from me either.

LOW and LOWEST have advantages too, but it just depends on the test. I've needed them before.

Don't let the words "high" and "low" confuse you. It's about speed, not necessarily quality. Scenario: Student in school taking test. Question #1: Who is the first President on the USA? Now the student can spend an hour (high setting) or a minute (low), and I can bet the answer will be the same (Washington). Run a 10-second test on a high-motion scene, and I bet you cannot tell the difference between ESTIMATE and HIGH/HIGHEST.

[supersam]

same here i use estimate search motion due to my testing that it yields the same results as high quality. i was doing high quality for a very long time then i found out that i can encode at a much faster rate using estimate. plus if ur using high bitrate then it really shouldnt be a problem using estimate. like txpharoah said play around with the settings and see what suits u best

[disturbed1]

Don't let the words "high" and "low" confuse you. It's about speed, not necessarily quality.
Run a 10-second test on a high-motion scene, and I bet you cannot tell the difference between ESTIMATE and HIGH/HIGHEST.

If your movie has the same consistancy of motion, like the God Father where most of the movie is people talking, estimate would be a good choice.

If your movie varies the amount of motion, like Star Wars, where there are equal parts high motion and no motion, estimate will not be a good choice.

Using motion estimate is just that an estimation. Which isn't always correct, it can't do correct DCT motion alignments of scene changes.

Using high quality takes longer to study each frame to compensate for it's level of motion and adjust the bitrate (accuracy) according to each frame, rather than a quick estimate.


10 seconds will not be a correct test. A better test would be to pick 10 different scenes each 2-3 minutes long with varied motion and complexity. Don't encode the scene seperately because the motion vectors used for each would change, but rather together as one clip. You could use DVD2AVI to cut out certain scenes, export the d2v files numbered movie-01, movie-02, etc. Use either VFAPI to convert all to avi (TMPG will load all numbered sequences) avisynth to join the d2v files, TMPG cut editing in source range.


Then study with a program like P-Frame which will show correct macro block placement and PSNR. Bitrate viewer can show relative Q level per gop if you only need a basic example.

[Will Hay]

Highest Quality (Very Slow) doesn't offer much improvement over High Quality (Slow) just doubles the encode time.

High Quality will yield better results than the others, with a good speed:quality ratio, much better than the speed:quality you get from Highest Quality setting.

Just set it to High Quality (Slow) and forget about it.

I changed the setting this morning as you suggested, and at lunch-time it had been running for nearly six hours yet the source position was only 42 mins (of 66 mins).
Much faster than the last try (very slow) but still too long
I think, as the footage I have is pretty much slow motion I'll go with estimate search next time and as you suggest, where I envisage there is high and low motion I'll go with high quality (slow).
Thanks for your input,
Will

[telemike]

What about using the "normal" setting?

[Will Hay]

What about using the "normal" setting?

Oh Christ, thanks for nothing!
Another option now, bugger!
You had to spoil it, didn't you?

Will

[energy80s]

I'm encoding a 62 minute Virtualdub capture with TMPGEnc using the following settings:
DVD PAL (MPEG-2 720x576 25fps VBR 7336kbps, Layer-2 48000Hz 384kbps)
...and I've changed the motion precision tab from motion estimate search (fast) to highest quality (very slow), just to ensure the highest quality I can get.
Trouble is, it's been encoding for almost ten hours and the source position is only 21 minutes!

This seems very slow. I have a 1600XP processor and have been encoding 4 minute DV clips to DVD with TMPGenc at highest quality settings and it takes about 35 mins. So for your 21 mins it should only take about 3 hours not 10! Sounds like you have some other issues going on there.

[Mac Everyman]

I agree with Energy80s -- make sure you don't have other programs running in the background slowing things down. I had the same problem until I shut down absolutely everything, including antivirus, firewall, etc. Improved my encoding speed about 20%.

[c3]

With 2.5GHz P4, encoding MPEG2 352x480 2200kbps with 2-pass VBR highest quality and default filtering takes about 16 hours for 1 hour of 720x480 AVI. It can take much longer if I use "high quality" filtering. Some features increase the encoding time dramatically.

[j1d10t]

Does it make much of a difference in motion search if you're using 2-pass? Won't the 2-pass find out where the motion is, and then set aside the correct bitrate for that scene?

[thete]

Yeah,. I would love to know that, as I am one of the daft ones who always uses highest quaility as well as a 2-pass, so it takens ages at times!

[IJM]

This looks like somethings wrong here.

I've checked the "Computer Details" in Silky31's profile and its almost the same as mine.

I'm running WinXP Pro on a XP2100+ with 512Mb RAM. I typically encode 30 minute clips to MPEG2 PAL 720x576 25fps VBR 4500kbps using 2-pass VBR and Highest Quality Motion Search. This is video only (I encode sound separately and mux it later), but that shouldn't make much difference. The only filters I include are usually to resize upwards and sometimes a bit of Noise Reduction. My source file is typically 29fps non-interlaced MPEG4 material.

Converting a 30 minute clip in this way usually takes me around 6 hours.

Are you using any additional filters (Noise Reduction etc.) because it seems to me like it's taking an unusually long time.

Ian.

P.S. Oh, I've just thought: I usually run Seti@Home (command line version), but I turn it off whilst encoding. You don't have anything else running do you?

[bkimmle]

I have a Athlon XP 2400+ and just encoded a 72 minute clip with it set the the Highest quality, 2 pass VBR, and a clip filter and it took 7 hours and 56 minutes. Do you have the cache set in TMPGEnc this really speeds the second half of the process up. I have my cache set to the harddrive I don't have the video on and set it to 5 Gig. This really helped out with VBR encodes.

[i_am_dave]

Silky,

By any chance, you don't have any filters turned on do you?
Filters usually double or triple encoding time in TMPGenc

[i_am_dave]

Then study with a program like P-Frame which will show correct macro block placement and PSNR. Bitrate viewer can show relative Q level per gop if you only need a basic example.

I have never heard of P-Frame. Sounds useful, can you give me more information?

[andie41]

As previously mentioned, any "extras" turned on can significantly increase render time - filters, motion settings, size, bitrate, etc.

I have an Athlon 1700XP slightly overclocked, running XP and generally encode 480x480, CBR 500000, 48hz, 224 kbits/sec. I also make sure to have my capture settings the same, so no extra conversion. Depending upon the video (lot of action or little action), I usually use the "motion estimate search" setting. With these settings, I get a 1.7 to 1 ratio.

I use CBR because I am not concerned about the file size and I have found that there is very little to gain in quality using a higher setting for motion search. at least for my videos. Occasionally I will use "normal". I find a 14% increase in time to go from "estimate" to " normal" and a 48% increase from "estimate" to "high quality." I haven't even bothered checking "highest".

Something that really increased the performance of my system was to go into "environmental setting" and under the cpu tab, check everything. SSE-2 is unavailable for my Athlon, but everything else, even the multi-thread options. Someone else posted this months ago, I tried it, and I get more than 10% decrease in render times with multi-threading enabled.

[RWANDREWS]

I am not qualified to talk about the various settings of normal vs highest quality, but I do use the slowest speed without any filters & I get about 4 1/2 hrs to encode 1 hour of AVI footage.

Buy the TMPG plus, it's worth it, as my encodes dropped about 30%. There are settings in the CPU environment that do not exist in the trial version. Make sure the boxes for SSE and MMX support are checked & set up a cache with at least 5GB of space. This does not seems to affect the estimated encode time, but makes the 2nd pass of VBR go about 3X faster. (I dont use CBR, except for VCD) I dont know what effect it may have on CQ or CBR encoding.

Hope that this helps.

[SaSi]

My experience with settings is the same. The highest quality is normally a waste. The only interesting about it is that if you encode using Constant Quality, the highest quality setting gives you a slightly smaller file for the same quality setting.

I have found that the quality setting is almost irrelevant for high bitrates. Silky used 7.3Mbps almost for average. This, I think, would give fine results even with the lowest quality which is damn fast.

My "favourite" setting for 1.5 ~ 2 hours DVD encoding is 4.8Mbps average bitrate with limits at 200kbps and 8Mbps with motion estimate in the fastest setting. It gives very acceptable speed and the image quality is fine (for my eyes).

The interesting thing is with captured VHS video. This is a bitch to encode. I guess that the noise in the content makes the encoder think the video has too much information. With such content, the fastest setting causes macroblocks boundaries to become quite visible at times.

[disturbed1]

Then study with a program like P-Frame which will show correct macro block placement and PSNR. Bitrate viewer can show relative Q level per gop if you only need a basic example.

I have never heard of P-Frame. Sounds useful, can you give me more information?

Teco Ltd sells a similar product http://www.tecoltd.com/fc.htm and

http://www.google.com/search?sourceid=navclient&ie=UTF-8&oe=UTF-8&q=p+frame+mpeg+analyzer

[Will Hay]

Hi folks,
Thanks for all the assistance.
No, I have no filters running although I did have Norton AV and Firewall running at the time of my first test.
I've decided to go with the 'slow' setting (rather than very slow), it's far, far faster than the 'very slow' option and as I type a 30 minute clip using...
...DVD PAL (MPEG-2 720x576 25fps VBR 8000kbps, Layer-2 48000Hz 384kbps is estimated to take three and a half hours, much more reasonable!
I've produced a DVD of the footage I based my initial test on and using the above settings it's superb.
Thanks again,
Will

[Will Hay]

These are my TMPGEnc settings, will they affect my encoding times you think?





I guess I should be ticking the cache setting in the CPU tab, right?
How do I save the (5gb?) cache to a drive other than the HDD the avi is stored on, and is this such a benefit?
Is it automatically deleted after encoding?
Thanks,
Will

[disturbed1]

Put a check in use multi thread, enable pipeling, enable pipeline for motion search, prefetch video, set it to 256mb (half your mem) leave check free space unchecked.

and enable save the results it will speed up the second pass. It uses as much drive space (in the temp file you set) as the encoded movie. If your shooting for a 4 gig movie it will take 4 gigs, 2 gig movie takes 2 gigs temp space.


You can also choose to edit bitrate by bps to get more accurate bitrate control. TMPG uses kbps (kilo bits) which can throw some calculations off.

[Will Hay]

Put a check in use multi thread, enable pipeling, enable pipeline for motion search, prefetch video, set it to 256mb (half your mem) leave check free space unchecked.

and enable save the results it will speed up the second pass. It uses as much drive space (in the temp file you set) as the encoded movie. If your shooting for a 4 gig movie it will take 4 gigs, 2 gig movie takes 2 gigs temp space.

Many thanks
Will

[i_am_dave]

deranged1,

Thanks for the links for the MPEG analyzers, interesting reading. I am waiting for a download login and password so I can look at the demo.

I have another question. Why do you select the CCIR-601 color space option?
Just curious, I have always used the default.

[disturbed1]

It seems to do red better, to eyes anyway.

deranged1 I like it.

[i_am_dave]

sorry about the deranged!

My E-bay name is deranged-1. I guess it was just some kind of Freudian thing.

[Will Hay]

My E-bay name is deranged-1.

I just checked your ebay bid history, you really should lay off those sex toys, haven't you enough already?
Will