Closed WhitePeter closed 2 weeks ago
Hi there,
Thanks for the issue report. We will look into this as soon as possible.
Hello! Thanks for writing in the bug report, you've got a great eye for spotting this! Will push the fix ASAP into the testing branch.
The following graphs are representations of the functions before/after. The y-axis represents the resulting QM value, and the x-axis represents the given base_qindex.
Previous, incorrect curve. The red line indicates the luma formula, and the orange line indicates the chroma formula. The chroma formula has since been removed due to the introduction of a separate qm param, chroma-qm-min, and its default of 8
.
New, updated and correct curve.
Interactive Desmos for learning how qm-min/max interact across qindexes.
For a given CRF before/after this patch, the bitrate may slightly rise (about a 6% increase, will likely vary depending on your source and CRF), though it does seem to improve detail as a result of both the formula change and the slightly increased allocation of bitrate.
If there are any unexpected results, feel free to let us know!
Ahh, gotta love open source. Went out for a walk and came back to find this addressed already, on a weekend no less. Great!
Your attention to detail is impressive, I'm surprised this was discovered in the first place! Thank you!
Right back at you. If it weren't for the detailed changelog I wouldn't have looked.
@WhitePeter Thanks a lot for the help :)
Originally posted by @WhitePeter in 119dab2
Please see above code comment. If I am mistaken just disregard my post.