How many images are used to train the v2 model?

EPFL-VILAB / omnidata

A Scalable Pipeline for Making Steerable Multi-Task Mid-Level Vision Datasets from 3D Scans [ICCV 2021]

Other

395 stars 49 forks source link

Open KexianHust opened 1 year ago

KexianHust commented 1 year ago

Great work! I know you added HM3D and 5 MiDaS datasets to the training data, but how many images in total were used to train the v2 model?

Twilight89 commented 1 year ago

Also have the same question, and how you preprocess the 5 MiDaS depth gt, also divided (2**16-1) like omnidata dataset?