facebookresearch / consistent_depth

We estimate dense, flicker-free, geometrically consistent depth from monocular video, for example hand-held cell phone video.
MIT License
1.61k stars 236 forks source link

How fast is the inference? #58

Open wtesler opened 2 years ago

wtesler commented 2 years ago

Before I get too far into this project, I'm interested in knowing how fast this works. Is it even remotely close to realtime? Or is it multiple seconds per frame?

Tetsujinfr commented 1 year ago

Although they show an animated gif with a smartphone "demo" of possible effects on the readme intro, lower on the readme they mention processing a 3s video takes 37mins on their 2080RTX system: haha good luck for realtime folks! the ML bs at its best!