README.md

Validation Datasets

We save all datasets into a unified format (uniformat) for easier testing. You can either download the datasets we pre-genenerated at this link, or you can download the original datasets and process them with the following steps.

Step 0:

Create a folder named datasets and put all datasets under it.

Step 1: download all raw datasets

Details provided later on this page.

Step 2: extract the uniformat datasets

Under the src folder, run sh robust_dc_protocol/save_uniformat_datasets_all.sh. This will save many npy files containing the rgb, sprse depth, and gt under datasets/uniformat_release.

Details on how to download the raw data

KITTI

Download the following files and unzip under the kitti_depth folder:

data_depth_annotated, data_depth_velodyne, data_depth_selection

Finally, download kitti raw images by:

cd datasets/kitti_depth
wget https://github.com/youmi-zym/CompletionFormer/files/12575038/kitti_archives_to_download.txt
wget -i kitti_archives_to_download.txt -P kitti_raw/
cd kitti_raw
unzip "*.zip"

The overall data directory is structured as follows:

kitti_depth
  ├──data_depth_annotated
  |     ├── train
  |     └── val
  ├── data_depth_velodyne
  |     ├── train
  |     └── val
  ├── data_depth_selection
  |     ├── test_depth_completion_anonymous
  |     |── test_depth_prediction_anonymous
  |     └── val_selection_cropped
  └── kitti_raw
        ├── 2011_09_26
        ├── 2011_09_28
        ├── 2011_09_29
        ├── 2011_09_30
        └── 2011_10_03

iBims-1

Download following its official link: https://www.asg.ed.tum.de/lmf/ibims1/

Unzip the file to get the following structure:

iBims
  ├──gt_depth
  │     ├── corridor_01.mat
  │     └── ...
  └──imagelist.txt

DIODE

We use the pre-processed images from marigold following this link:

Unzip the tar file under datasets/marigold and you will get

marigold
  └──diode
      ├── indoors
      │      ├──scene_00019
      │      └──...
      └── outdoors
             ├──scene_00022
             └──...

ETH3D

First download follow this link by marigold:

Unzip the tar file under datasets/marigold and you will get

marigold
  └──eth3d
      ├──depth
      │   ├── courtyard_dslr_depth
      │   └──...
      └──rgb
          ├── courtyard
          └──...

To extract the sparse depth from COLMAP, download the COLMAP scenes from the ETH3d website. Specifically, follow this link to get the COLMAP models for the distorted rgbs.

Put the downloaded files under datasets/eth3d_raw and get:

eth3d_raw
  ├──courtyard
  │   └── dslr_calibration_jpg # the colmap model
  └──...

Finally, run sh write_eth3d_colmap_sparse_depth_all.sh under this src/robust_dc_protocol folder, which will write the sparse depth maps under the marigold folder:

marigold
  └──eth3d
      ├──depth
      │   ├── courtyard_dslr_depth
      │   └──...
      ├──rgb
      │   ├── courtyard
      │   └──...
      └──sparse_depth
          ├── courtyard
          └──...

ARKitScenes

Follow the official instruction here. You only need the Validation split of the depth upsampling subset.

Expect something like this:

ARKitScenes
  └──depth
      └──upsampling
          ├──Validation
          │    ├──41069021
          │    └──...
          └──metadata.csv

VOID

First download the zip files (you can use gdown) under datasets:

cd datasets
https://drive.google.com/open?id=1rzTFD35OCxMIguxLDcBxuIdhh5T2G7h4

Under the datasets folder, unzip the downloaded file and you will get:

void_release
    ├── void_150
    │    ├── data
    │    │     ├── birthplace_of_internet
    │    │     └── ...
    │    ├── test_absolute_pose.txt      
    │    └── ...
    ├── void_500
    │    └── ...
    └── void_1500
         └── ...

NYUv2

We used preprocessed NYUv2 HDF5 dataset provided by Fangchang Ma.

cd datasets
wget http://datasets.lids.mit.edu/sparse-to-dense/data/nyudepthv2.tar.gz
tar -xvf nyudepthv2.tar.gz

After that, you will get a data structure as follows:

nyudepthv2_h5
  ├── train
  │    ├── basement_0001a
  │   ...   ├── 00001.h5
  │         └── ...
  └── val
       └── official
            ├── 00001.h5
            └── ...