Tech Journal Back to Tech Journal

What's the difference between the Divx ;-) Fast-Motion and Low-Motion CODECs?

This is written as an observer of both video CODECs used. Fast-Motion (FM) and Low-Motion (LM) are very similiar pieces of code, with significant (but few) differences between the actual algorithems.

In my test case, Episode 13 of Neon Genesis Evangelion, I copied the VCD digitally (in a computer) and the used the CODECs on the resultant MPEG-1 file. I got the folowing:

CODEC Output AVI size Video Format
Fast-Motion 55.7MB (58,447,872 bytes) 40 KBps
Low-Motion 134MB (140,736,512 bytes) 97 KBps

All other information not directly associated with the above was identicle, including the paramaters given to the CODEC within each of their configuration dialog:
    Keyframe every 10 seconds
    Compression Control = 100 (most crisp)
    Data Rate = 910 kilobits/se

Just by looking at the above, you can see that LM used the bandwidth given to it more efficiently (to the hilt!) -- using 97 of a maximum of 113 KBps. Of course, the maximum KBps during playback was probably higher (not observed). It seems that LM saw more scenes where there was a need for the data, while FM did not. This probably stems from the fact that the video was (for the most part) a long series of nearly still (static) scenes, each lasting from 1 second to 5 seconds.

Wacthing the encoding process, I saw that in any fairly static scene FM would invest nearly 0K (always less than 1K), while LM would use between 3K to 10K. The only time LM would 'agree' to use 0K on a scene was when the scene was totally static (example: during a static white caption on a totally black screen).

Watching the video, you can see that any motion faster than a few dozen pixels across the screen was handled beautifully by FM (sometimes better than the rest of the scene), while LM mangled it. On the other hand, in any scene where part of the picture stopped, the MPEG blocks-blob effect would occur. If you had a cynlinder with (origionally) shaded colors from left to right, this was displayed perfectly by LM, while FM turned the shades into concentric near-circles (blobs) of same-color, which danced and shifted in the regular MPEG the-pixels-aren't-moving-but-the-color/picture-is display issue which existed mostly in MPEG-1.

In summary, any nearly-still scene looked better in LM, while any kind of note-worthy motion was handled better by FM. (i.e. the CODECs do what they say they'll do: hand fast- and low- motion.) Since the episode was mostly still scenes, LM worked better overall, at the cost of being nearly 2.5 times larger. The way I see it, FM find high-motion scenes more facinating, while LM holds low-motion scenes in high regard. Each will dedicate more bandwidth to its personal interest, and try to make the scenes it like to be as perfect as possible.

As for myself, I will do backups using both, since even at the cost of 200MB each, it still comes to less than the original 251MB MPEG-1 file, while getting it in a format that is much more usable (but not as portable), because of its size.

Last updated on 2001-02-01 14:00:00 -0700, by Shalom Craimer

Back to Tech Journal