Monocular surface normal maps for the Cityscapes fine-annotated
leftImg8bit frames, predicted with Lotus
(a diffusion-based dense prediction model).
Each normal map corresponds 1:1 to an image in the Cityscapes gtFine/leftImg8bit fine-annotation split.
This repository contains only the predicted normal maps, not the RGB images — pair them with your own
copy of leftImg8bit from the official Cityscapes download page.
Note: only train and val are… See the full description on the dataset page:
https://huggingface.co/datasets/Sta8is/cityscapes_lotus_normals.
