Unfortunately, the output video is stretched vertically. It seems like
filter crops around 60px from the bottom of original video, then scales
1860x1080 to 640x360. Curious that the very first frame is scaled correctly.
Scale_npp function doesn't have this problem, but I would prefer use
On Thu, 21 Mar 2019 at 05:25, Александр via ffmpeg-user <
[hidden email]> wrote:
> Hello, I am trying to do full hardware video transcoding with scaling.
> Here is my console command:
> ffmpeg -hwaccel nvdec -hwaccel_output_format cuda -i "input.mp4" -vf
> scale_cuda=-2:360 -c:v h264_nvenc -b:v 2000k "output.mp4"
> Input stream info:
> Stream #0:0(und): Video: h264 (High) (avc1 / 0x31637661), yuv420p,
> 1920x1080 [SAR 1:1 DAR 16:9], 1711 kb/s, 1 fps, 1 tbr, 16384 tbn, 2 tbc
> Output stream:
> Stream #0:0(und): Video: h264 (h264_nvenc) (Main) (avc1 / 0x31637661),
> cuda, 640x360 [SAR 1:1 DAR 16:9], q=-1--1, 2000 kb/s, 1 fps, 16384 tbn,
> 1 tbc (default)
> Unfortunately, the output video is stretched vertically. It seems like
> filter crops around 60px from the bottom of original video, then scales
> 1860x1080 to 640x360. Curious that the very first frame is scaled
> Scale_npp function doesn't have this problem, but I would prefer use
> Could you help me to make this filter work as expected?
Unfortunately, the scale_cuda parameters are quite limited, as you can
ffmpeg -h filter=scale_cuda
Compared to scale_npp:
ffmpeg -h filter=scale_npp
Is there a specific reason as to why you'd prefer scale_cuda over
I wouldn't want to make the assumption that scale_cuda is analogous to the
scale filter, wherein you'd be able to use specific flags that the latter
> Is there a specific reason as to why you'd prefer scale_cuda over
Indeed, scale_cuda accepts width and height options only. In my
situation I have a small project on Windows and as far as I know, there
is no way to compile static version of ffmpeg with libnpp support. That
is why scale_cuda is more convenient to me, because it doesn't require
CUDA SDK installed.
Even so, the problem remains and I would be very grateful if someone
more experienced than me would confirm this issue before reporting it.
This filter was added almost 2 years ago, I believe more in my
incompetence than in existence of such a critical and obvious bug which
corrupts the output.
ffmpeg-user mailing list
[hidden email] https://ffmpeg.org/mailman/listinfo/ffmpeg-user
To unsubscribe, visit link above, or email
[hidden email] with subject "unsubscribe".