Hello, I'm trying to transcode couple camera streams on low-end i3 box equipped with nvidia gpu, but even with latest git version, I'm not able to make it work. My goal is to do decoding, resizing and encoding w/o copying frames to/from ram, according to https://trac.ffmpeg.org/wiki/HWAccelIntro it should be working if I use cmd like this: ffmpeg -hwaccel_device 0 -hwaccel cuvid -c:v h264_cuvid -i 'udp://@239.1.1.1:5555?fifo_size=1000000&overrun_nonfatal=1&buffer_size=100000' -vf scale_npp=-1:360 -vcodec h264_nvenc -preset slow -b:v 1200k -bufsize 500k -minrate 1200k -maxrate 1200k -f mpegts -y 'udp://239.2.2.2:10001' But as soon as I add resize, I get "green" screen effect - <https://www.dropbox.com/s/cvutg01i6crn4nv/Screenshot%202017-03-21%2004.09.24.png?dl=0> http://imgur.com/WljkD6g without resize filter - it's working just fine. I've also tried using solution from https://trac.ffmpeg.org/ticket/5587 but even if I found pixel format which is accepted, results are the same. I believe that problem is somewhere in pixel format, but can't figure out what I'm missing :(