/NeRFBK

NeRFBK: Different datasets for evaluating NeRF-based methods for 3D reconstruction

NERFBK: A HOLISTIC DATASET FOR BENCHMARKING NERF-BASED 3D RECONSTRUCTION

This is a repository of image collections - called NeRFBK - with real and synthetic data specifically assembled to support researchers in evaluating and comparing the performances of NeRF algorithms against a reliable and accurate Ground Truth (GT).

The data contains:

  • industrial objects (reflective and textureless metallic surfaces);
  • transparent objects (mainly glasses);
  • heritage scenarios (small and large scales, including lost scenes);
  • large scale urban areas acquired with drone or airborne cameras.

Researchers can compare NeRF methods on textured, textureless, metallic, transparent and aerial scenes to optimize and validate techniques for real-world use, such as in industrial inspections, cultural heritage preservation or large-scale urban 3D modeling. For each dataset, GT data vary from LiDAR or Terrestrial Laser Scanning to high-resolution photogrammetry. Using these GT, quantitative evaluations can be performed to provide an unbiased comparison of geometric accuracy. Different approaches and metrics can include best plane fitting, cloud-to-cloud comparison, profiling, accuracy and completeness analyses, RMSE, etc. For each scenario, images and GT data can be downloaded from the given links.

If you use these data and you publish any result, please acknolwedge this repository and its related publications. Thanks!


Related Papers


Datasets - Downloading available

Dataset Numb. images Camera type &
image size
Approx.
size(cm)
Description Ground Truth
(GT)
I
N
D
U
S
T
R
I
A
L
Industrial_A 295
(ca 200 MB)
Huawei p20 pro,
1080x1920 px
5x5x4 Textureless,
Small and Complex,
Reflective,
Video
Triangulation laser scanner
(ca 171 MB)
Industrial_B 220 Huawei p20 pro,
1920x1080 px
15x12x4 Textureless,
Complex,
Reflective,
Video
Triangulation laser scanner
(ca 190 MB)
Synthetic 373
(ca 250 MB)
Virtual pinhole camera,
1920x1080 px
11x11x2 Well-textured,
Sharp edges,
Video
Synthetic data
(ca 1 MB)
Synthetic_Metallic 300
(ca 750 MB)
Virtual pinhole camera,
1080x1920 px
16x16x13 Textureless,
Complex,
Reflective,
Video
Synthetic data
(ca 10 MB)
T
R
A
N
S
P
A
R
E
N
T

and

R
E
F
L
E
C
T
I
V
E
Glass 552
(ca 380 MB)
Huawei p20pro,
1920x1080 px
5x5x25 Complex transparent shape,
Highly refractive,
Video
Photogrammetry on powdered surface
(ca 50 MB)
Cup 287 Huawei p20pro,
1080x1920 px
8x8x10 Complex shape,
Highly refractive,
Video
Photogrammetry
(ca 138 MB)
Bottle 300
(ca 200 MB)
Huawei p20 pro,
1080x1920 px
6x6x30 Complex shape,
Highly refractive,
Video
Photogrammetry on powdered surface
(ca 290 MB)
Synthetic_Glass 300
(ca 750 MB)
Virtual pinhole camera,
1080x1920 px
10x10x22 Transparent,
Highly refractive,
Video
Synthetic data
(ca 6 MB)
H
E
R
I
T
A
G
E
Doss Trento 761
(ca 150 MB)
Huawei p20pro,
540x960 px
2500x2500
x1500
Outdoor large scale Terrestrial Laser Scanner
(ca 3 GB)
Cyprus monument 178
(ca 100 MB)
Samsung S6,
3840x2160 px
1500x1500
x500
Outdoor large scale, video Photogrammetry with Reflex camera
(ca 500 MB)
Statue 100
(ca 100 MB)
Ideal reflex camera,
6016x4016 px
200x100
x500
Outdoor large scale,
Two scales
Synthetic data
(ca 2 MB)
Vase 50
(ca 1 GB)
Google Pixel2,
4024x3016 px
40x30 Indoor Photogrammetry with Reflex camera
(ca 500 MB)
Metopa 106
(ca 500 MB)
Nikon COOLPIX P90,
4000x3000 px
100x80x10 Indoor museal detailed relief Terrestrial Laser Scanner
(ca 800 MB)
Tunnel 4353 RealSense D455,
640x480 px
8000x300
x200
Underground,
Mobile robot based,
High frame-rate
Terrestrial Laser Scanner
(ca 200 MB)
Neptune temple 680
+
214
Nikon D3X,
6048x4032 px,
Canon EOS 550D,
5184x3456 px
2500x1500
x7000
Terrestrial
+
UAV images
Outdoor large scale
Terrestrial Laser Scanner
(ca 50 MB)
Trento Duomo 565
(ca 4 GB)
Samsung S6,
5312x2988 px
8000x8000 Outdoor large scale,
Terrestrial
Photogrammetry with Reflex camera
(ca 2.6 GB)
Modena cathedral 132
+
58
Nikon D750,
6016x4016 px
Canon EOS 600D,
5184x3456 px
2500x2500 Terrestrial
+
UAV images
Large scale architecture
Terrestrial Laser Scanner
(ca 4.8 GB)
Baalshamin temple 47
(ca 100 MB)
Multiple cameras,
From 600x399 px
to
4288x2848 px
500x1500 Lost object,
Sub-optimal baselines,
Unordered touristic images,
Multiple resolutions
-
A
E
R
I
A
L
Drone / UAV 224
(ca 6 GB)
Sony,
7952x5304 px
City scale Outdoor large scale,
Built-up and vegetated areas
Airborne Laser Scanner
(ca 3 GB)
Dortmund 59
cam0
(ca 2GB),
cam1
(ca 1.8 GB),
cam2
(ca 1 GB),
cam3
(ca 1 GB),
cam4
(ca 1 GB)
IGI Multi-sensor, 5 views
6132x8176 px at Nadir, 50 mm
8176x6132 px at Oblique, 80 mm
City scale Outdoor large scale,
Built-up and vegetated areas
Airborne Laser Scanner
(ca 220 MB)

NeRFBK was created by the 3D Optical Metrology unit (3DOM) of Fondazione Bruno Kessler (FBK).


License

The data provided here is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.


Changelog

  • [15 Jul, 2023] Uploaded all datasets
  • [26 Jul, 2023] Text and links fixing
  • [20 Oct, 2023] Link to papers added