Closed quamt closed 5 months ago
"fp16" is written wrong and should be "all", so --vpp-nlmeans fp16=<x> should take "none", "blockdiff" and "all".
I'll fix the help in next release.
Online document has be fixed for --vpp-nlmeans fp16.
Thank you for your feedback. I will now mark this issue as resolved and close it.
Hello @rigaya,
I'm currently using the
--vpp-nlmeans
feature and have some questions regarding thefp16
parameter options. While integrating this into my workflow, I’ve run into a few confusing points concerning the parameter values and their documentation.Documentation on
fp16
Settings: According to the documentation, the options forfp16
include:none
: Usesfp32
for higher precision but slower performanceblockdiff
: Usesfp16
in block difference calculations for a balance between performance and precisionfp16
: Usesfp16
in weight calculations for faster performance but lower precisionEncountered Errors: When I set the
fp16
parameter using--vpp-nlmeans fp16=fp16
, I receive the following error:Interestingly, when I try
fp16=auto
, which the error message suggests as a valid option, I get a similar error. This option (auto) isn’t listed in the initial documentation.Confusion with Parameter Values:
It appears that the valid options listed in the error messages do not entirely match what’s in the documentation. Also, while
none
andblockdiff
work as expected, they are not mentioned in the error message as valid values.Thanks a lot for your help and the awesome work on this tool!_