Definitive video codec?

I ve finally switched to W7 and I am therefore tuning my machine, I ve just regained the functionality of Text(ex9) node and now I am facing my first gig with W7. I am rendering some animations with blender as plain jpg files so I will create videos out of them with Virtualdub, what is the best option for encoding-decoding-codecs ?

I’ve tried a bunch of codecs, but I haven’t found anything that works better than Indeo Video 5.10 with forced keyframe every frame. This is what I use for serious video scrubbing.

However, I should note that while I run my patch on a high-end W7 workstation I do most video encoding and VVVV testing on an old Thinkpad T60p Windows XP laptop.

Tx, so about Virtualdub, I can just load the still frame and save as AVI with default settings? Or anything to change?

m-jpeg (motion jpeg) codec is recommanded usually…

for my part i m using or AVI raw ( if triggering inside the video) or WMV classical codec.

really strange,m-jpeg never work out in my case,the Indeo Video 5.10 pretty good and wmv.

yes, mjpeg.

i use picvideo mjpeg quite often, with quality set to 18 or 19
decoding is done with ffdshow with picture settings luminance correction and fullrange in the videodecoder settings.

what works too is playing a dds image sequence directly from a SSD. this changes the full workload to reading speed of the ssd and leaves the cpu pretty much free for anything else.

looks like the debate is open… could everybody also leave the computer specs and in particular the OS version ?

ive had really good results throughout w-xp (atom processor!!), vista, w7 with cinepack which comes preinstalled and ofcourse forced 1keyframe per frame. For vjing its ok as i dont need bitdepth and resolution at its most, but ive read on that the have developed their own codec (free to use) that i havnt tried yet but the best of is that the decoding gets done in the actual gpu which sounds exiting )

the first video i ve rendered is about 4Gb for about 1 minute. Now, I guess that s a bit too big a file to be played smoothly on vvvv, so, in virtual dub, which settings to you advice to get a properly playable file? Or do you advice any other program to get a video out of the still frames?

don’t want to destroy your proudness of resolume visualex, but that codec is some copy of regular mjpeg codec (don’t remeber witch one), done pretty bad btw. And GPU decoding, has nothing to do with a codec, it’s only a matter of player if it supports GPU or not

for now u7angel’s way is only reliable

as for the rest
beware - indeo, had couplle sample lib’s died for no reason, opens ok but jams everyhing in middle.

cinipak - never had list “normal” result with it

general - don’t use different codecs, framerate dies if you mix such samples.

antokhio: No it’s not some mjpeg derivative, it uses DXT1 compression. see e.g. comments here
Why "done bad"ly, could you elaborate/provide sources for that?

Yes, GPU decoding has to do with codec. Without DXT1 or similar gpu-compatible compression method, decompression could not be efficiently done on the GPU.
Wikipedia: Unlike some image compression algorithms (e.g. JPEG), S3TC’s fixed-rate data compression coupled with the single memory access (cf. Color Cell Compression and some VQ-based schemes) made it ideally suited for use in compressing textures in hardware accelerated 3D computer graphics.

One caveat is that you get GPU decompression (and the performance gain associated with it) only when using Resolume Avenue.

sounds like playing DXT compressed frames from hdd, which works also great for me… best solution up to a certain resolution. there is also image-cacher which might help…

they also kinda-sorta offered to help in extending gpu accel to other software, but so far it seems no one has taken them up on the offer
If any other VJ software is interesed in supporting it then we’d love to hear from them. (in above link)

i did quite extensive testing after using picvideo for some time to check if there isn’t anything better to use with vvvv filestream.

i tested nasty gradients, vector-style graphics and normal videostuff with all kinds of codec. wmv, h264 with ffdshow, some gpu decoding special stuff and so on.

in the end picvideo mjepg proved to be the most accurate and cpu friendly one. but this all might change once vux gets his new quicktime compatible video node ready to go.

out of interest: why? what do you expect from a quicktime layer?

It also turned out in other places (e.g. vjforums) that (picvideo) mjpeg is the best codec for VJ-like work (random access, scratching, quality, cpu&hdd load). nowadays, at least for avenue, dxv is a probably better alternative.

Still up for answer/tips, I got this Blender project (1300x800), I can render still frames and encode them as video lately or do a video in Blender directly, I am a bit lost on how to encode it so that I can run it nicely on vvvv, I don t need scrubbing, just play the videos, (a few minutes each).

well maybe quicktime is being easier to handle, because i guess the plugin is threaded and therefore uses multicores by default. hence less interuption of vvvv. and as its not using directshow, no wait for frame and sync business.

aaand quicktime h264 just works and delivers good quality, although new tests need to be executed.

If there is no need for scrubbing I´d use Xvid.
Encode it with the official codec and play it with ffdshow tryouts.

I think vux mentioned something about Quicktime being buggy and therefore he wanted to move to vlc/libav…