metadata
dataset_info:
features:
- name: image
dtype: image
- name: conditioning_image
dtype: image
- name: text
dtype: string
splits:
- name: train
num_bytes: 60051214963
num_examples: 118285
download_size: 60053644708
dataset_size: 60051214963
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
I cropped the Coco 2017 dataset to 512 and created a depth image with DepthAnything v2. https://cocodataset.org/#download