My Frameserve with AVISynth is slower? :S

[reload_effect]

hi

i have been using tmpgenc for all my conversions of vhs to vcd, and i thought i would give frameserving a go to try and speed up my conversions, so i had a read of the avisynth web site and made up a script from their guide to frameserve my file.

i capture to an AVI file with MJPEG compression of 19 using VirtualVCR with a setting of YUY2, however avisynth says it isn't YUY2

My system specs.
Duron 1ghz
Windows XP
512mb Ram
Leadtek Winfast 2000xp capture card

here is my script, for basically just deinterlace and resize the file

AviSource("d:\capture.avi", pixel_type="YUY2")
ConvertToYV12() # if necessary
Trim(0, 1544)
Telecide()
LanczosResize(352, 288)

i tested with a 1 minute file with 3 programs, tmpgenc, mainconcept and CCE, here were my results

Tmpgenc Normal convert (no frameserve, TMPGEnc set to High Quality and deinterlaces and resizes) - 2 Minutes 11 seconds

Tmpgenc Frameserve (only setting is quality set to high) - 2 Minutes 31 seconds

Mainconcept - 1 Minute 42 Seconds

CCE - 1 Minute 24 Seconds

Mainconcept and CCE were just a trial to check the quality because i was thinking of changing products to speed up my conversions, i used a few guides from the guide section of this site, however the quality was worse than TMPGEnc, which was more than likely because i have the wrong settings as i have read that it can be the same on here.

I know that there was only 20 seconds difference, but that was for only 1 minute of encoding and i normally do 3-4 hours of source encoding a day.

i was expecting avisynth frameserving to be faster, and i didn't, well i dont think i did, add any noise filters etc to the file, i thought it was just a straight deinterlace and resize.

if you know of anything i have done wrong, or can help me make it quicker, without a quality drop, i'm all ears

thanks

[DJRumpy]

You don't need to specify the pixel type, and the parameter sets the output format, not the input format. Also, the newest versions of AVISynth (2.5+) output YV12 natively, there is no reason to place the ConvertTo parameter there.

Last but not least, you should not use LanczosResize when reducing to VCD or SVCD formats. It produces very sharp lines, which are difficult for an MPEG encoder to compress. You would be better served by BilinearResize, or even BicubicResize if you still want some sharpness. LanczosResize is usually fine for CVD and DVD formats, where bitrate isn't at such a premium cost.

As to your frameserving speed question, frameserving an AVI to an encoder will almost always ( I try never to say never ) be slower than simply encoding the AVI itself. Think about it. In the first case, your opening the AVI in another piece of software, which is taking each frame, modifying it as you requested, and then handing it off to your encoder. If you just encode the AVI directly, the encoder can work directly on each frame, without having to wait on the frameserver. The only exception I can think of is the fact the TMPGenc works in the RGB colorspace, which is very slow. AVISynth can work in YUY2, or YV12, which is much faster. If you were doing alot of modifications, then frameserving might actually increase your speed over TMPGenc alone.

Frameserving is intended for those not wanting to create an entire AVI to encode. Most of us start out that way. You use something like VirtualDub to edit your avi, and save it as a new avi, which takes hours. Frameserving simply lets you skip the 'save new AVI' step, and modify the AVI on the fly, while feeding it to the encoder. TMPGenc includes many AVI editing capabilities, which means frameserving to it is typically, entirely optional. Most people who frameserve are using newer (read faster) encoders.

Encoded properly, you can expect about the same quality from all three encoders, but TMPGenc seems to suffer a bit more from macroblocking on VCD and SVCD formats than Main Concept, or CCE.

[reload_effect]

wow. thanks for the great reply, you brought up quite a few points i didn't even think about.

i would love to use CCE or Mainconcept for the speed of both pieces of software, i'll have to try and find a guide for creating the highest quality possible vcd/test and figure them out for myself

thanks again

[DJRumpy]

If your doing VCD, the settings are pretty simple. Under the VIDEO settings, find the Quantization button. Click it and set it for Ultra Low Bitrate, then click OK. You really don't need any other options on the VIDEO page. Audio, you would set for whatever you need. Under the Quality page, I would disable the Low Pass Filter, the Effect Restricted Vertical Filter, and Dithered Quantization. For the 'Quantizer Characteristics, set the slider to 32.

Set your Intra DC Block Precision to 8, and set your Block Scanning Order to ZigZag. Last but not least, place a check next to the 'Progressive Frame Flag' setting. These settings assume a progressive source...if your source is interlaced, everything still applies, but your Block Scanning should be set to Alternate, and your Progressive Frame Flag should be disabled.

On the main properties page, set your MODE to MPEG-1 SS (System Stream), and set your aspect ratio to 4:3.

[Ejoc]

Of corse its slower if you are using telecide(). Comment it out at give it a try, it will be much faster. I would be surprised if telecide() was slower then the internal IVTC of tmpgenc, I tried it before and when it wasnt crashing it was slower then telecide().

I would use:

AviSource("d:\capture.avi")
Trim(0, 1544)
Telecide() #only if you have IVTC issues
LanczosResize(352, 288)

I use AVISynth all the time, and its faster, if all I need to do is resize my video I still run it thru avisynth as it reduces my encoding time. And I favor LanczosResize().

[reload_effect]

thanks, i'll give both of those methods a try out this afternoon

[reload_effect]

sorry for the late reply

i tried both those methods

- there was no difference in the speed between deinterlacing with avisynth and tmpgenc

- i tried the settings for CCE, however tmpgenc still had superior quality.

thanks