musicvid10 wrote on 7/6/2011, 11:32 AM
I took 20 intermediate codec candidates that can be rendered directly in Vegas, both RGB and YUV, 8-bit and 10-bit, and did a shootout using but one main criteria -- how well will they do on the Belle-Nuit chart, which shows primaries, resolution, and chroma subsampling comparisons. In each case I used 720p renders of the 720p version of the chart, because that's the highest either my notebook or desktop monitor will resolve.
I found eight 4:4:4 codecs that reproduced the chart (8 bit RGB) measure-for-measure (save for the 1-bit shift):
-- AVI Uncompressed
-- MOV Uncompressed
-- MOV BMP
-- MOV PNG
-- HuffYUV (RGB mode)
-- Lagarith (RGB mode)
-- UT (RGB mode)
[Magic YUV, which came after these tests, also produces a clean chroma subsmpling image, and will be included in subsequent tests.
Some "visually lossless" codecs did not make it into this first round, because they either showed shifts in color primaries or chroma subsampling when compared to the seven above, whose results are apparently identical in every respect. Among these were Cineform, DNxHD, Matrox, and Animation.
Knowing that render times and file sizes would be entirely irrelevant on static material, I then rendered a three-minute video from each, the relative render times and file sizes appearing in the chart above.
As can be seen, UT gives the best render time, with Lagarith or UT being the best compromise in both areas.
Here are the things I did NOT do:
-- Attempt to identify which codecs will work for a particular purpose or application in real-world conditions
-- Run multiple generations to determine where rounding errors begin to show
-- Attempt to determine how levels and gamma would behave in external applications
So even though they are lossless at various degrees of efficiency, don't assume that your favorite external encoder (x264) or NLE is going to play nice.
Judging differences in the next round of "visually lossless" intermediates gets even tougher, because some codecs fudge with primaries, show chroma subsampling errors, or both in varying ways and degrees.
These tests were all run in 2011 in Vegas Pro 8 on a dual-core laptop, nothing fancy.