CRF prediction

[Sharc]

@jdobbs
For BD5/720p the initial CRF is 11.0 (way too low), and then steps up to 30.5 (way too high). Is this large initial swing intentional? I am asking because typical final CRF values are in the range 18 ..... 24 for a BD5/720p, and it takes currently typically 6...8 steps to converge.

[jdobbs]

@jdobbs
For BD5/720p the initial CRF is 11.0 (way too low), and then steps up to 30.5 (way too high). Is this large initial swing intentional? I am asking because typical final CRF values are in the range 18 ..... 24 for a BD5/720p, and it takes currently typically 6...8 steps to converge.If you could give me the disc name and your post settings I could test it and see why it does it. But since the CRF function will act completely differently depending on the disc and settings, I'd need more information to debug it.

As for why it starts low... I did testing early on and established the starting number... but I haven't looked at it in a long time.

[Sharc]

It does it for the vast majority of disks for my typical settings:
- Target Size BD-5
- Movie only
- Resize to 720p
- 2 Audio AC3 480 or 640 kbps
- 1 or 2 subtitles
- 1 pass CRF
- High Quality (Default)

Example:
[17:56:36] BD Rebuilder v0.38.09 (beta)
- Source: THEKINGSSPEECH
- Input BD size: 19.65 GB
- Approximate total content: [01:58:14.629]
- Target BD size: 4.39 GB
- Windows Version: 6.0 [6002]
- MOVIE-ONLY mode enabled
- RESIZE 1080p to 720p enabled
- Quality: High Quality (Default), CRF
- X264 Tweak(s) enabled (--tune film --me umh --subme 8)
- Audio Settings: AC3=1 DTS=1 HD=0 Kbs=448
[17:56:38] PHASE ONE, Encoding
- [17:56:38] Extracting A/V streams [VID_00005]
- [18:07:43] Reencoding: VID_00005 (1 of 2)
- [18:07:43] Collecting video information
- Source Video: MPEG-4 (AVC), 1920x1080
- Rate/Length: 23.976fps, 170'077 frames
- Convert: 1280x720, 23.976fps, 170'077 frames
- [18:07:43] Performing CRF Prediction...
- Analyzing 11.00 30.50 25.68 22.41 20.51 19.60 19.23 [19.10]
- [18:30:14] Encoding using constant rate factor.
- [22:42:30] Video Encode complete
- [22:42:30] Reencoding audio tracks (if req'd)
- [22:51:44] Multiplexing M2TS
- [22:54:50] Extracting A/V streams [VID_00000]
- [22:54:51] Reencoding: VID_00000 (2 of 2)
- [22:54:51] Collecting video information
- Source Video: MPEG-4 (AVC), 1920x1080
- Rate/Length: 23.976fps, 24 frames
- Convert: 1280x720, 23.976fps, 24 frames
- [22:54:51] Reencoding: VID_00000, Pass 1 of 1
- [22:54:53] Video Encode complete
- [22:54:53] Multiplexing M2TS
[22:54:54]PHASE ONE complete
[22:54:54]PHASE TWO - Rebuild Started
- [22:54:54] Rebuilding AVCHD file Structure
[22:59:17] - Encode and Rebuild complete
[22:59:17]JOB: THEKINGSSPEECH finished.

[Sharc]

@jdobbs
Did you have a chance to look into this (http://forum.doom9.org/showthread.php?p=1626347#post1626347)?

Another minor issue:
Downscaling to 1440x1080 uses spline16resize as Default filter, while the Default filter should be Bicubic according to the documentation.

[jdobbs]

@jdobbs
Did you have a chance to look into this (http://forum.doom9.org/showthread.php?p=1626347#post1626347)?Sorry, I'd forgotten about that one.Another minor issue:
Downscaling to 1440x1080 uses spline16resize as Default filter, while the Default filter should be Bicubic according to the documentation.Hmm... I must have missed that one, I'll check it out.

[jdobbs]

@jdobbs
Did you have a chance to look into this (http://forum.doom9.org/showthread.php?p=1626347#post1626347)?

Another minor issue:
Downscaling to 1440x1080 uses spline16resize as Default filter, while the Default filter should be Bicubic according to the documentation. I fixed both of these for the next release. I also made a fix for the issue that was reported some time back in which a quick-hit could cause a final adjustment that was way off in CRF prediction.

[Sharc]

I fixed both of these for the next release. I also made a fix for the issue that was reported some time back in which a quick-hit could cause a final adjustment that was way off in CRF prediction.
Great! Thank you.