search

LLaVA-VL / LLaVA-NeXT

Apache License 2.0
2.4k stars 167 forks source link

Llava-Next-OV FPS #172

Open ehayeshaiper opened 3 weeks ago

ehayeshaiper commented 3 weeks ago

Hi, thanks for open sourcing your excellent work. Can I ask what FPS the video training is done at? In your video inference example you uniformly sample 16 frames, which works out to be slightly less than 1 per second.

Edit: Just found it in the paper, should we always run inference at 1 FPS?

Luodian commented 3 weeks ago

You can refer the sampling logic here, the fps in default to 1

https://github.com/LLaVA-VL/LLaVA-NeXT/blob/e98849102929e1c6304b60b28cca541567b7b643/llava/utils.py#L25