search

facebookresearch / consistent_depth

We estimate dense, flicker-free, geometrically consistent depth from monocular video, for example hand-held cell phone video.
MIT License
1.63k stars 236 forks source link

Support for MobileNetV3 or similar to run on mobile devices #7

Closed dblatner closed 4 years ago

dblatner commented 4 years ago

Is it possible to run the model on a mobile device (iOS, Android) or compatible with neural architectures like MobileNetV3, EfficientNet-B0/1/3 ?

I like to see the performance and accuracy in action on real-time captured video streams from the device.

roxanneluo commented 4 years ago

Hi, currently the method is offline and is not realtime. To adapt to these models, it won't be too hard to implement an interface similar to the MiDaS one and enable MobileNetV3 as the backend.