H.264 All-Intra Data Rates Significantly Higher

does built-in h.264 codec encode i-frame files differently? i trying determine optimal gop length high bitrate exports. image quality seems degrade, in i-frames, when using key frame distances greater one.

 

after performing series of tests characterize adobe h.264 encoder, discovered exported files larger when key frame distance equals 1 frame (n=1). average video data rate test file rendered adobe h.264 encoder follows:

 

n=1 : 2.17 bpp : 24i

n=2 : 0.66 bpp : 12i + 12p

n=3 : 0.59 bpp : 8i + 8b + 8p

 

note how data rate drops 70% (from 2.17 0.66 bpp) though 50% of i-frames still exist when n=2. comparison, here video data rate when exporting quicktime h.264:

 

n=1 : 0.89 bpp : 24i

n=2 : 0.70 bpp : 12i + 12p

n=3 : 0.64 bpp : 8i + 16p

 

the following chart shows data rates @ key frame distances 1-48 frames adobe h.264, quicktime h.264 (via adobe), quicktime pro, , expression encoder 4 pro. data rates consistent among encoders @ gop lengths greater one. there anomoly adobe h.264 codec compressing all-intra files.

 

the observed behavior occurs in profiles, tested @ levels 4.1, 4.2, 5.0, , 5.1:

 

image quality better in adobe h.264 all-intraframe file, respect detail retention. pictures below show sections of 2 consecutive frames magnified 400%. file n=2 less accurate , contains noticeable blocking. i-frames don't in files n>1.

 

the test file seventeen second premiere pro sequence consisting of h.264, mpeg-2, , ae files effects applied. exports rendered premiere pro timeline , v210 uncompressed 4:2:2 intermediate file of sequence. following settings used:

 

format: h.264

width: 1280

height: 720

frame rate: 24 fps

field order: progressive

aspect: square pixels (1.0)

tv standard: ntsc

profile: baseline, main, , high

levels: 4.1, 4.2, 5.0, 5.1

render @ maximum bit depth: enabled

bitrate encoding: vbr, 2-pass

target bitrate: maximized each profile/level

maximum bitrate: maximized each profile/level

key frame distance: 1, 2, 3, 4, 5, 6, 7, 8, 12, 24, 48

use maximum render quality: enabled

multiplexer: mp4

stream capability: standard

 

software:

 

adobe media encoder cs6 creative cloud v6.0.3.1 (64-bit)

premiere pro cs6 creative cloud v6.0.3

windows 7 sp1

quicktime pro windows v7.6.9 (1680.9)

mediainfo 0.7.62 (for gop , data rate information)

i think short answer yes,

a h.264 encoder i-frame differently. frames complete expressions of picture no temporal compression information.

p frames use _p_redictive information. ie information prior frames.

b frames use _b_i-directional predictive frame information.

 

h.264 gets majority of it's bit saving use of b , p frames. when i-frame only block compression , none of advantages of p , b frames. gop n=1 doesn't bit's per pixel.

 

having said find comment...

profitic wrote:

 

note how data rate drops 70% (from 2.17 0.66 bpp) though 50% of i-frames still exist when n=2. comparison, here video data rate when exporting quicktime h.264:

 

... interesting. indeed, why datarate 70% less when should @ best 50% gop n=2. 50% less should same i-frame information plus 0 bytes b frame between them. (gop = i,b,i). more , throwing away bits i-frame. so, seems to ratecontrol bug.

 

 

http://en.wikipedia.org/wiki/group_of_pictures

 

"the gop structure referred 2 numbers, example, m=3, n=12. first number tells distance between 2 anchor frames (i or p): gop size. second 1 tells distance between 2 full images (i-frames): gop length. example m=3, n=12, gop structure ibbpbbpbbpbbi. instead of m parameter maximal count of b-frames between 2 consecutive anchor frames can used."

More discussions in Adobe Media Encoder (AME)

adobe