Trying to deinterlace with Bob NVEnc; job crashes

[antoniu200]

Rectification after testing once again.

-resize works just fine, but looking at my schedule, I probably won't have the time to properly test it soon. Who knows, maybe I'll be able to test it in December, but until then, I won't.

What I can tell you for now is that a command such as

ffmpeg -c:v mpeg2_cuvid -resize 1280x720 -deint 2 -i "D:\Video\EDIT.mp4" -c:v h264_nvenc -profile:v high -rc vbr -cq 18 -g 500 -r 50 -vf setsar=1:1 -c:a copy

works just fine, without errors.
Maybe last time I tried it with -resize after -deint? I don't remember, but the cmd.zip should lighten me up regarding this.

The input file in this command is 720x576.

EDIT: Last time, I tried -crop, not -resize. My mistake.

[Selur]

Question was wether:
a. crop, resize, deint all work
b. wether their order is important (different results? some orders not working) or whether ffmpeg internally always uses the same order.

Cu Selur

[antoniu200]

So, I come with some news once again. This time it might solve the libnpp dependency issue.

By pure luck, I came across this: 
https://stackoverflow.com/questions/4187...in-windows. Here I saw that the CUDA SDK includes all needed NPP libraries to compile FFmpeg with -enable-libnpp. If needed, there is also a step-by-step guide on how to compile it. I don't know how to compile custom project, I never had the push to try that. If you do know, however, how to do that and try what it says there, that would be great, as that would mean the CUVID resize option would be fully functional and could potentially provide more flexible ways of resizing.

Since the other post I liked to 2 months ago mentioned some licensing and legal issues, I read the CUDA SDK EULA. I am no lawyer, but it clearly states:

2.6. Attachment A
The following CUDA Toolkit files may be distributed with Licensee Applications developed by you, including certain variations of these files that have version number or architecture specific information embedded in the file name - as an example only, for release version 9.0 of the 64-bit Windows software, the file cudart64_90.dll is redistributable.

Which basically means anyone can build an app and distribute it using the libs in the table, along which NPP is included:

Component NVIDIA Performance Primitives Library
Windows nppc.dll, nppc.lib, nppial.dll, nppial.lib, nppicc.dll, nppicc.lib, nppicom.dll, nppicom.lib, nppidei.dll, nppidei.lib, nppif.dll, nppif.lib, nppig.dll, nppig.lib, nppim.dll, nppim.lib, nppist.dll, nppist.lib, nppisu.dll, nppisu.lib, nppitc.dll, nppitc.lib, npps.dll, npps.lib
Mac OSX libnppc.dylib, libnppc_static.a, libnppial.dylib, libnppial_static.a, libnppicc.dylib, libnppicc_static.a, libnppicom.dylib, libnppicom_static.a, libnppidei.dylib, libnppidei_static.a, libnppif.dylib, libnppif_static.a, libnppig.dylib, libnppig_static.a, libnppim.dylib, libnppisu_static.a, libnppitc.dylib, libnppitc_static.a, libnpps.dylib, libnpps_static.a
Linux libnppc.so, libnppc_static.a, libnppial.so, libnppial_static.a, libnppicc.so, libnppicc_static.a, libnppicom.so, libnppicom_static.a, libnppidei.so, libnppidei_static.a, libnppif.so, libnppif_static.a libnppig.so, libnppig_static.a, libnppim.so, libnppim_static.a, libnppist.so, libnppist_static.a, libnppisu.so, libnppisu_static.a, libnppitc.so libnppitc_static.a, libnpps.so, libnpps_static.a

I suggest you read this EULA as well, if you're ok with trying to build FFmpeg with -enable-libnpp, as this one is not very long.

[Selur]

Building ffmpeg with libnpp for Windows should be relatively easy since https://github.com/m-ab-s/media-autobuild_suite alredy supports it.
But before adding it to Windows I would also:
a. need an alternative to https://evermeet.cx/ffmpeg/ for MacOS binaries which supports it
b. need someone to figure out how to modify https://github.com/Selur/HybridDeploy to also build them on Linux
Since I really have no motivation to look into it and there's probably nobody who is, so it's probably not happening.

Cu Selur