Video frame extraction throughput

How to run

• python benchmarks/run.py --write
• Or: PIXELTABLE_HOME=$(mktemp -d) python benchmarks/scenarios/video_frames.py --clips 1 --out /tmp/vid.json --scale reduced

Results

1 clip, 108 frames, 0.24 s wall, 0.21 s CPU, 450 frames per wall second.

Clips

1

Frames

108

Insert

0.24 s wall, 0.21 s CPU

Frames per wall second

450

Methods

Scale of this page

reduced: 1 clip

Clip

scripts/release-bench/fixtures/sample.mp4

Iterator

frame_iterator() every frame, plus a 64x36 thumbnail column

Machine

Intel(R) Xeon(R) Processor, 4 cores, 15.6 GB, Linux 6.12.94+, Python 3.12.3, pixeltable 0.7.21

Recorded

2026-10-28, command python benchmarks/run.py --write

Limitations

• One short clip. A production library is longer files, more codecs, and disk that is not this VM.
• Each step ran once, with no warmup, so every time is one sample. scripts/release-bench repeats its runs.
• The thumbnail is 64x36. A larger resize would change the 450 frames-per-second figure.
• Do not compare this number to the /engineering video-to-frames series without reading both methods. That series uses CPU time across releases on a Mac.

Where Pixeltable fits

If you only need ffmpeg on the command line, Pixeltable is the wrong tool. Use it when each still has to stay tied to the source clip and later columns (thumbnails, detectors, an index).

import pixeltable as pxt
from pixeltable.functions.video import frame_iterator

TableModel = pxt.model_base()

class Videos(TableModel, name='videos'):
    video: pxt.Video

class Frames(
    TableModel,
    name='frames',
    base=Videos,
    iterator=frame_iterator(video=Videos.video),
):
    thumb = frame.resize((64, 36))

Questions