File size: 9,157 Bytes
0190190
 
 
 
 
 
3fc9fef
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0190190
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
---

language:
- en
size_categories:
- 10K<n<100K
---

# Asset Inspection Dataset

This dataset's scenes are arranged as follows:

  ```
  scene/
  ├── cam_parameters.tar
  ├── depths.tar
  ├── images.tar
  ├── scene.glb Scene 
  └── scene.blend
  ```

The office building scene has four surface soiling settings, so it is arranged as follows

  ```
  office_building
  ├── cam_parameters.tar
  ├── depths.tar
  ├── high_soiling_images.tar
  ├── low_soiling_images.tar
  ├── medium_soiling_images.tar
  ├── office_building.glb
  ├── very_low_soiling_images.tar
  └── Office_model.blend
  ```

## Images

`images.tar` files and `*_images.tar` contain the images in the dataset. Noise _has_ been added - refer to the paper for details on our additive noise model. The images in the bridge and crane scenes are .pngs and office building images are .jpgs to save space.

## Depth Maps

`depths` contains the depth maps. These are .exr files which store the metric depth. Each pixel has a three values each of which is the depth. To open a .exr file, one can use the [OpenEXR](https://openexr.com/en/latest/) python library or the image viewer [tev](https://github.com/Tom94/tev). A function for opening .exr files using OpenEXR in python is included below :

``` python
import OpenEXR

def open_exr(f_name: str):
    with OpenEXR.File(f_name) as infile:
        header = infile.header()
        channels = infile.channels()
        keys = list(channels.keys())
        if keys == ["RGB"]:
            img = channels["RGB"].pixels
        if keys == ["RGBA"]:
            img = channels["RGBA"].pixels
        elif keys == ["X", "Y", "Z"]:
            X = channels["X"].pixels
            Y = channels["Y"].pixels
            Z = channels["Z"].pixels

            height, width = Z.shape

            img = np.empty((height, width, 3))
            img[..., 0] = X
            img[..., 1] = Y
            img[..., 2] = Z

    return img, header
```

Each pixel of the depth map contains the metric depth in each of the three channels. Therefore, if we return to the previous code and `img[..., 0]` will get the depth for each pixel in the image.

## Camera Intrinsics and Extrinsics

Camera intrinsics and extrinsics for each frame are included in `cam_parameters`. Each frame number has an individual .json file which is formatted as follows:

``` json
{
 "loc": [
  world_x_position,
  world_y_position,
  world_z_position
 ],
 "rot": [
    w,
    x,
    y,
    z
 ],
 "focal_length": f_mm,
 "sensor_width": sensor_width_mm,
 "clip_end": max_dist_m,
 "clip_start": min_dist_m
}
```

The `loc` field refers to the position of the camera in the blender world coordinate frame. The `rot` field refers to the rotation quaternion in the blender world coordinate frame (in scalar first format (w, x, y, z)). `focal_length` refers to the focal length of the camera in millimetres. `sensor_width` refers to the width of the camera sensor in mm. `clip_start` and `clip_end` are the minimum and maximum distance settings that blender renders.

To convert these coordinates into the familiar COLMAP coordinate system, one can use the following lines of code:

``` python
  from scipy.spatial.transform import Rotation

colmap_rot = [rot[1], rot[0],
              rot[3], -rot[2]]
colmap_transform = Rotation.from_quat(colmap_rot, scalar_first=True)
T_vec = -colmap_transform.apply(loc)
```

## Mesh model

For each scene, we exported a mesh model. To produce a ground truth point cloud, we sampled the mesh using [Open3D](https://www.open3d.org/). To replicate this, one can use:

```python
import open3d as o3d

mesh = o3d.io.read_triangle_mesh(glb_file_path)
pointcloud_sampled = mesh.sample_points_uniformly(
    number_of_points=1000000)
```

## Blender Scenes

These blender scenes were constructed in [Blender 5.0.1](https://www.blender.org/) from assets from [BlenderKit](https://www.blenderkit.com). To render the office building with different soiling settings, use the `render_office.py` script which can be done using Blender's in built scripting capability or with the following shell command:

``` bash
blender -b Office_model.blend --python render_office.py 
```

### Sources for Blender Assets

We used the following assets in our dataset (some of them substantially modified):

1. Mirazev, A.: Model: Modern Building. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/bab4567f-7f71-44f7-af6b-96f5d60cb66e/>, last accessed 09 Mar 2026
2. Jacinto, M., Pinto, J., Patrikar, J., Keller, J., Cunha, R., Scherer, S., Pascoal, A.: Pegasus Simulator: An Isaac Sim Framework for Multiple Aerial Vehicles Simulation. In: 2024 International Conference on Unmanned Aircraft Systems (ICUAS), pp. 1–8. IEEE (2024). <https://doi.org/10.1109/ICUAS60882.2024.10556959>
3. Se-Cam, J.: Model: 80s Cartoon Wagon Car. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/5150ca16-f5c0-49e9-b2a4-bdf9d6a51e1a/>, last accessed 09 Mar 2026
4. Se-Cam, J.: Model: Cartoon Car / Toy Car. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/b24f0fab-da30-4a85-bdec-c3b2fced3bb7/>, last accessed 09 Mar 2026
5. Se-Cam, J.: Model: 80s Cartoon Pick-Up. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/c587fce1-af5a-4f77-9e78-af10526cf4a5/>, last accessed 09 Mar 2026
6. Tirindelli, D.: Model: 6kw Solar Panels Structure. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/ba8093b7-416c-461c-af22-ac311df86b03/>, last accessed 09 Mar 2026
7. Rexodus: Model: Antenna. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/c67c9184-3ebd-4809-adf5-91b1dd0ae759/>, last accessed 09 Mar 2026
8. Mitroi, R.: Model: AC Unit Big. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/1a06752c-d093-474a-b870-a011deb2366e/>, last accessed 09 Mar 2026
9. Wells, C.: Material: Procedural Concrete Tiles. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/3b90f722-6d01-4f62-be56-f392e9312bac/>, last accessed 09 Mar 2026
10. Paludo, L.: Material: Simple Concrete Procedural. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/6156591b-e7a4-4aee-a6cc-0380517e4003/>, last accessed 09 Mar 2026
11. Share Textures: Material: Dark Steel. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/f170e0bf-f784-482e-86eb-1d7e307a22cc/>, last accessed 09 Mar 2026
12. Michaud, A.: Material: Textured Painted Wall (Procedural). BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/9891df1f-b429-461a-bc5e-b0520e25a120/>, last accessed 09 Mar 2026
13. Yang, R.: Material: White Wall 02. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/9891df1f-b429-461a-bc5e-b0520e25a120/>, last accessed 09 Mar 2026
14. Adhe, E.: Material: Black Painted Plaster Wall. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/61c2c84d-1701-4b19-9728-f0710bab1cce/>, last accessed 09 Mar 2026
15. Pambudi, R.R.: Material: Material Window Glass Cycle. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/4367fa4b-9633-48a2-862f-224fbf7b1730/>, last accessed 09 Mar 2026
16. Yang, R.: HDRi: Urban City Highsky Bluesky. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/139e9edb-ba7f-4848-b6c5-3b5c227ca805/>, last accessed 09 Mar 2026
asset-gallery-detail/aa82da45-2e29-4e4d-9ec4-631977b4d29b/>
17. Cosmo: Model: Suspension/Arch Bridge. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/5358ddd3-6dbf-4a1c-8731-4c6bdcfa8338/>, last accessed 09 Mar 2026
18. MapacheDRelease: Model: Crane Tower. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/eb5c137e-007e-485b-bbb9-675e76fcdac9/>, last accessed 09 Mar 2026
19. Mohammadi, S.: Material: Realistic Procedural Grass (Cycles/Eevee). BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/a465bd1e-a14d-442f-90b4-0bfb02dffa9f/>, last accessed 09 Mar 2026
20. Godoi, M.: Material: Procedural Concrete. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/6312d40c-7afb-49e3-b167-c772147ce40d/>, last accessed 09 Mar 2026
21. Russo 3D: Material: Procedural Brick. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/f5ab3780-1489-4c62-8050-b400bbd14e95/>, last accessed 09 Mar 2026
22. Middleton, J.: Material: Bricks Procedural Wall. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/fa2d62e6-fc17-4310-a0bf-b35eaefcfd3c/>, last accessed 09 Mar 2026
23. Middleton, J.: Material: Brick Wall Procedural. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/79f3e321-4923-45dd-8508-8082163db23d/>, last accessed 09 Mar 2026
24. FreePoly: HDRi: Over The Construction Site Yellow Mud. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/66411e15-4a1f-46cb-8e3a-0271e83dc8e4/>, last accessed 09 Mar 2026
25. FreePoly: HDRi: Lake Highsky Blue Sky. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/aa82da45-2e29-4e4d-9ec4-631977b4d29b/>, last accessed 09 Mar 2026
26. NK Productions: Material: Asphalt Road Procedural. BlenderKit, <https://www.blenderkit.com/asset-gallery-detail/0f2ec127-987c-49ab-9478-7f777bdde2d3/>, last accessed 09 Mar 2026

NB. [2] was used for it's drone model.