-
Notifications
You must be signed in to change notification settings - Fork 8
Time consumption of ffio
dongrixinyu edited this page Jan 25, 2024
·
1 revision
- aiming to test the speed of Python wrapper for FFmpeg-C-API compared with pure FFmpeg-C-API
- The test condition:
- AMD x86_64 (16 cpu cores),
- Ubuntu 18.04, gcc, FFmpeg-6.0, Python 3.10.
- video: 29.97fps, 1920*1080p, 2M/s
- running 1 loop of decode-ffio is composed of decode-C(purely call FFmpeg-C-API to get 1 frame) and decode-py-convertor(convert C image frame to python format,
numpydefault). Same to encoder items. - The table below shows that decoding 1 frame costs
15.84msin average, in which1.49msis used to convert decoded image in C format(unsigned char *) to python format(numpy, etc.). encoding 1 frame costs11.96msin average, in which2.03msis used to do C-py format conversion.
| functions | average cost time(ms) | max cost time(ms) | min cost time(ms) |
|---|---|---|---|
| decode-ffio | 15.84 | 42.28 | 8.08 |
| decode-C | 14.34 | 40.43 | 7.00 |
| decode-py-convertor | 1.49 | 2.63 | 1.02 |
| encode-ffio | 11.96 | 22.65 | 10.37 |
| encode-C | 9.92 | 18.41 | 9.13 |
| encode-py-convertor | 2.03 | 4.49 | 1.05 |