performance problem about transform360 HOT 10 OPEN

The processing frame rate is proportional to resolution. What are the width/height?

from transform360.

resolution is 4k

from transform360.

input is 4k. output is 720p.I think threaddata is the key. Your filter cannot utilize all cpu core to do the job, or this is a limit because transform cannot do in multi thread?

from transform360.

Multithread is not supported yes. This project is not designed for supporting realtime applications.

That said, I just transformed a stereo 4K x 4K 60FPS equirect video to a 1080x1440 stereo cubemap with 17fps processing (0.277x) on my mac book. Perhaps something other than filtering is your bottleneck.

from transform360.

Input #0, flv, from 'vr.flv':
  Metadata:
    make            : Insta360
    model           : Insta360 Pro
    encoder         : Lavf57.66.102
  Duration: 00:10:00.04, start: 0.000000, bitrate: 6193 kb/s
    Stream #0:0: Audio: mp3, 48000 Hz, mono, s16p, 64 kb/s
    Stream #0:1: Video: flv1, yuv420p, 3840x2160, 200 kb/s, 29.97 fps, 29.97 tbr
, 1k tbn

from transform360.

sample clip:
https://iloveyaya.ga:9443/test/tmp/vr_sample.flv
for 30 sec

from transform360.

conversion command:
ffmpeg -i vr.flv -vf transform_v1="input_stereo_format=MONO:w_subdivisions=4:h_subdivisions=4:max_cube_edge_length=512:w=1280:h=720" -c:v libx264 -b:v 2000k -t 30 -c:a aac -ar 44100 -b:a 64k -f flv out.flv

from transform360.

With your file and command I got 20fps processing (0.651x) on my 8-core mac book.

from transform360.

I run it on win8.1 x64 platform.
Because vs2015 compile ERROR against macro SOFTWARE_PREFETCH_OPT
I comment out "#define SOFTWARE_PREFETCH_OPT" to build.

from transform360.

I modify config to "input_stereo_format=MONO:w_subdivisions=4:h_subdivisions=4:max_cube_edge_length=512"

Got 0.629x speed on linux test machine.

from transform360.