Doom9's Forum - View Single Post - Taking the dumbness out of CCE multipass

Kedirekin · 25th April 2004, 02:36

Jeez. Where did the day go. I didn't get the lawn mowed (it started raining here too) and I only got the oil drained (but that will at least force me to finish the job tomorrow).

Well, at least I managed to try your analysis trick, despite some completely baffling issues opening avs files in CCE.

I encoded Pokemon 5 (it was handy) OPV with Q of 28, plugged the numbers into the KBPS_frm_FileSize spreadsheet (thank god for Open Office), and it told me I'd need a bitrate of 3884. That's quite a bit higher than I expected for that Q level - almost twice what I expected in fact.

I think I'll try encoding it OPV with 28 and see how it turns out.

Some other things that have occurred to me:

you'd probably want to match your SelectRangeEvery to your GOP size (1200,12 for 3/4, 1500,15 for 3/5). Doing otherwise would probably induce CCE to introduce too many scene change detections, distorting the analysis
have you thought about the effect of scene changes that happen to fall in these 12 frame samplets? Do you think they would match the affect of scene changes in the full encode? I find this a bit of a mind twister - do 10 short GOPs in a 1% sample have the same effect as 1000 short GOPs in a full encode?
Heck, what effect will 12 frame samplets have. I believe each samplet will result in a closed GOP, which will again distort analysis (not sure how much though)
The samplets are ½ second, with 50 seconds between them. Not sure where I'm going with this, but I have a nagging feeling in the back of my mind that there is some fairly common scenario where the samplets will miss something, and produce missleading results.

BTW: we have different definitions of wasted effort. I'll gladly let my PC encode overnight if it will save me 10 minutes of work time. The cost of my personal free time figures heavily in my gestalt.

25th April 2004, 02:36	#7 \| Link
Kedirekin Registered User Join Date: Oct 2001 Location: Minnesota Posts: 1,110	Jeez. Where did the day go. I didn't get the lawn mowed (it started raining here too) and I only got the oil drained (but that will at least force me to finish the job tomorrow). Well, at least I managed to try your analysis trick, despite some completely baffling issues opening avs files in CCE. I encoded Pokemon 5 (it was handy) OPV with Q of 28, plugged the numbers into the KBPS_frm_FileSize spreadsheet (thank god for Open Office), and it told me I'd need a bitrate of 3884. That's quite a bit higher than I expected for that Q level - almost twice what I expected in fact. I think I'll try encoding it OPV with 28 and see how it turns out. Some other things that have occurred to me: you'd probably want to match your SelectRangeEvery to your GOP size (1200,12 for 3/4, 1500,15 for 3/5). Doing otherwise would probably induce CCE to introduce too many scene change detections, distorting the analysis have you thought about the effect of scene changes that happen to fall in these 12 frame samplets? Do you think they would match the affect of scene changes in the full encode? I find this a bit of a mind twister - do 10 short GOPs in a 1% sample have the same effect as 1000 short GOPs in a full encode? Heck, what effect will 12 frame samplets have. I believe each samplet will result in a closed GOP, which will again distort analysis (not sure how much though) The samplets are ½ second, with 50 seconds between them. Not sure where I'm going with this, but I have a nagging feeling in the back of my mind that there is some fairly common scenario where the samplets will miss something, and produce missleading results. BTW: we have different definitions of wasted effort. I'll gladly let my PC encode overnight if it will save me 10 minutes of work time. The cost of my personal free time figures heavily in my gestalt.