View Full Version : B-frames --> File not smaller?
ChAoS Overlord
1st April 2002, 12:13
So far I've done 7 encodes of monthy python episodes with DivX5. I only used b-frames from the pro options, now my filesizes are EXACTLY as gordian knot predicts them to me (using divx4 bitrate calc). I thought that b-frames greatly reduced filesize? Why doesn't it do this with my encodes? They all turn out to be exactly 233-234MB each.
BTW: I use the gain Pro codec, with gain disabled using msconfig & removed the web3000 reg entries with adaware. (win98 SE)
domingol
1st April 2002, 12:34
I'm getting exactly the same here. It isn't the matter what tools were used (or without tools), the size is always as GKnot predicted it'll be. And i'm confused too, cause i've heard a lot of people getting undersized files with B-frames enabled, when i'm getting lots of 700 MBs rips since i use divx5 :confused:
P.S. Sorry for my poor english :D
SpEeDaMiGo
1st April 2002, 16:24
As far as I know DivX5 takes into account the "space" it saves by using B-frames and it "fills up this space" to reach the desired average bitrate. So IMHO using B-frames theoretically shouldn't lead to undersized files. The undersized-problem was already common with DivX4 and now when using the standard DivX5 (which really hasn't changed much) it still occurs. Using B-frames then even increases the tendency of DivX4/5 to make undersized files, since it can save up a lot of space.
The main reason for this prob IMHO is that people choose too low resolutions, ie too high %-values in the comp checks (have a look at all this threads like "GKnot gave 1.4GB, I only got 1GB) and as a result the codec is "saturated". I've already made quite many DivX5 rips and with 700MB rips I always end up in +/- 2MB. 1400MB rips quite often stay between 1390MB and 1400MB.
I always make a compressibility check with GKnot and DivX3 and try to aim for 45%-55% when using DivX5 with B-frames (also depends a lot on the kind of movie .. if I got a feeling movie xy has difficult or critical or "eye-catching" scenes, I'll go higher or whatever ... can't tell absolute values ... just some guideline). Those are just my thoughts and "likings" about the percent-values. People may like completely different ones, but for ME they work well.
So, as a rule of thumb: Choose high resolutions and neutral or sharp bicubic resize and no noise filter (most DVD sources are quite clean). I only apply some tempsmoother (level 1 or 2) if the movie is really grainy and has "grainy noise".
Furthermore, wait with the audio encoding and fill up the "undersized" space with MP3/Ogg data. And if you really got a stubborn movie why not add the AC3 track or a second language ... just some ideas and my thoughts.
Hope, this helped a little.
Feel free to correct me or tell me your opinion.
Doom9
1st April 2002, 19:34
if you're doing 2 pass b frames should have no impact on size whatsoever.. do a test using CQ mode as I did (news of last month) and you'll see the impact. Though, with divx5 the size mostly changes since every b frame gets its quantizer doubled.
ChAoS Overlord
1st April 2002, 20:11
Originally posted by Doom9
if you're doing 2 pass b frames should have no impact on size whatsoever
But I reckon quality should improve, no?
domingol
1st April 2002, 21:14
Yes, the quality really improves at low motion scenes, but on high motion, the effects of the blocks are very noticeable, mainly if you have (like me) a P3@500, and canīt go higher than 1 at post-processing without constant freezes.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.