search

apm1467 / videocr

Extract hardcoded subtitles from videos using machine learning
MIT License
506 stars 117 forks source link

speed increase by (optionally) skipping (most) frames #48

Open drzraf opened 1 year ago

drzraf commented 1 year ago

Text is generally aimed at human reading thus usually lasting ~ 1 second or so (rarely less than 500ms)

We can increase speed by 10 or 15 by skipping frames. Said otherwise only OCR'ing one frame every (fps/2)