dataset.md

Data download and preparation

Download the scenes (We use Brandenburg gate, Trevi fountain, and Sacre coeur in our experiments) from Image Matching Challenge PhotoTourism (IMC-PT) 2020 dataset

Download the train/test split from NeRF-W and put under each scene's folder (the same level as the "dense" folder, see more details in Tree structure of each dataset

Run

ROOT_DIR="/mnt/cephfs/dataset/NVS/nerfInWild/brandenburg_gate/"
img_downscale=2
python prepare_phototourism.py --root_dir $ROOT_DIR --img_downscale $img_downscale
#$ROOT_DIR is the directory of dataset
#$img_downscale is an integer, e.g. 2 means half the image sizes

to prepare the training data, This will largely accelerate the speed of data preparation step before training. The generated data will be saved on the same level as the "dense" folder

Tree structure of each dataset

brandenburg_gate/
├── dense/
│   ├── images/
│   ├── sparse/
│   │      ├──depth_maps/
│   │      ├──depth_maps_clean_300_th_0.10/
│   ├── stereo/
│   
├── cache/
│ 
├──brandenburg.tsv


trevi_fountain/
├── dense/
│   ├── images/
│   ├── sparse/
│   │      ├──depth_maps/
│   │      ├──depth_maps_clean_300_th_0.10/
│   ├── stereo/
│   
├── cache/
│ 
├──trevi.tsv


sacre_coeur/
├── dense/
│   ├── images/
│   ├── sparse/
│   │      ├──depth_maps/
│   │      ├──depth_maps_clean_300_th_0.10/
│   ├── stereo/
│   
├── cache/
│ 
├──sacre.tsv