EPFL-VILAB / omnidata

A Scalable Pipeline for Making Steerable Multi-Task Mid-Level Vision Datasets from 3D Scans [ICCV 2021]
Other
418 stars 53 forks source link

Unflatten workaround #26

Closed bileamScheuvens closed 1 year ago

bileamScheuvens commented 2 years ago

Experimenting with midas on torch 1.12.1 led to a well documented but unfixed issue concerning jit tracing and unflatten layers (https://github.com/pytorch/pytorch/issues/49538). This is a workaround from the mentioned thread (as well as minor adjustments respecting warnings) that should otherwise have no impact.

alexsax commented 1 year ago

Thanks!