JDihlmann commited on
Commit
1ecdaee
·
1 Parent(s): 062ceb1

feat: Readme

Browse files
Files changed (3) hide show
  1. README.md +150 -0
  2. other/dataset.png +3 -0
  3. other/preprocessing.png +3 -0
README.md ADDED
@@ -0,0 +1,150 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 🕯️ Light-Stage OLAT Subsurface-Scattering Dataset
2
+
3
+ *Companion data for the paper **"Subsurface Scattering for 3D Gaussian Splatting"***
4
+
5
+ > **This README documents *only the dataset*.**
6
+ > A separate repo covers the training / rendering **code**: <https://github.com/cgtuebingen/SSS-GS>
7
+
8
+ <p align="center">
9
+ <img src="other/dataset.png" width="80%" alt="Dataset overview"/>
10
+ </p>
11
+
12
+ ## Overview
13
+
14
+ Subsurface scattering (SSS) gives translucent materials (wax, soap, jade, skin) their distinctive soft glow. Our paper introduces **SSS-GS**, the first 3D Gaussian-Splatting framework that *jointly* reconstructs shape, BRDF and volumetric SSS while running at real-time framerates. Training such a model requires dense **multi-view ⇄ multi-light OLAT** data.
15
+
16
+ This dataset delivers exactly that:
17
+
18
+ * **25 objects** – 20 captured on a physical light-stage, 5 rendered in a synthetic stage
19
+ * **> 37k images** (≈ 1 TB raw / ≈ 30 GB processed) with **known camera & light poses**
20
+ * Ready-to-use JSON transform files compatible with NeRF & 3D GS toolchains
21
+ * Processed to 800 px images + masks; **raw 16 MP capture** available on request
22
+
23
+ ### Applications
24
+
25
+ * Research on SSS, inverse-rendering, radiance-field relighting, differentiable shading
26
+ * Benchmarking OLAT pipelines or light-stage calibration
27
+ * Teaching datasets for photometric 3D reconstruction
28
+
29
+ ## Quick Start
30
+
31
+ ```bash
32
+ # Download and extract one real-world object
33
+ curl -L https://…/real_world/candle.tar | tar -x
34
+ ```
35
+
36
+ ## Directory Layout
37
+ ```
38
+ dataset_root/
39
+ ├── real_world/ # Captured objects (processed, ready to train)
40
+ │ └── <object>.tar # Each tar = one object (≈ 4–8 GB)
41
+ └── synthetic/ # Procedurally rendered objects
42
+ ├── <object>_full/ # full-resolution (800 px)
43
+ └── <object>_small/ # 256 px "quick-train" version
44
+ ```
45
+
46
+ ### Inside a **real-world** tar
47
+ ```
48
+ <object>/
49
+ ├── resized/ # θ_φ_board_i.png (≈ 800 × 650 px)
50
+ ├── transforms_train.json # (train-set only) ⇄ camera / light metadata
51
+ ├── transforms_test.json # (test-set only) ⇄ camera / light metadata
52
+ ├── light_positions.json # all θ_φ_board_i → (x,y,z)
53
+ ├── exclude_list.json # bad views (lens flare, matting error, …)
54
+ └── cam_lights_aligned.png # sanity-check visualisation
55
+ ```
56
+ *Raw capture* Full-resolution, unprocessed RGB-bayer images (~ 1 TB per object) are kept offline—contact us to arrange transfer.
57
+
58
+ ### Inside a **synthetic** object folder
59
+ ```
60
+ <object>_full/
61
+ ├── <object>.blend # Blender scene with 112 HDR stage lights
62
+ ├── train/ # r_<cam>_l_<light>.png (= 800 × 800 px)
63
+ ├── test/ # r_<cam>_l_<light>.png (= 800 × 800 px)
64
+ ├── eval/ # only in "_small" subsets
65
+ ├── transforms_train.json # (train-set only) ⇄ camera / light metadata
66
+ └── transforms_test.json # (test-set only) ⇄ camera / light metadata
67
+ ```
68
+ The *small* variant differs only in image resolution & optional `eval/`.
69
+
70
+ ## Data Collection
71
+
72
+ ### Real-World Subset
73
+
74
+ **Capture Setup:**
75
+ - **Stage**: 4 m diameter light-stage with 167 individually addressable LEDs
76
+ - **Camera**: FLIR Oryx 12 MP with 35 mm F-mount, motorized turntable & vertical rail
77
+ - **Processing**: COLMAP SfM, automatic masking (SAM + ViTMatte), resize → PNG
78
+
79
+ | Objects | Avg. Views | Lights/View | Resolution | Masks |
80
+ |---------|------------|-------------|------------|-------|
81
+ | 20 | 158 | 167 | 800×650 px | α-mattes |
82
+
83
+ <p align="center">
84
+ <img src="other/preprocessing.png" width="60%" alt="Preprocessing pipeline"/>
85
+ </p>
86
+
87
+ ### Synthetic Subset
88
+
89
+ **Rendering Setup:**
90
+ - **Models**: Stanford 3D Scans and BlenderKit
91
+ - **Renderer**: Blender Cycles with spectral SSS (Principled BSDF)
92
+ - **Lights**: 112 positions (7 rings × 16), 200 test cameras on NeRF spiral path
93
+
94
+ | Variant | Images | Views × Lights | Resolution | Notes |
95
+ |---------|--------|----------------|------------|-------|
96
+ | _full | 11,200 | 100 × 112 | 800² | Filmic tonemapping |
97
+ | _small | 1,500 | 15 × 100 | 256² | Quick prototyping |
98
+
99
+ ## File & Naming Conventions
100
+ * **Real images** `theta_<θ>_phi_<φ>_board_<id>.png`
101
+ *θ, φ* in degrees; *board* 0-195 indexes the LED PCBs.
102
+ * **Synthetic images** `r_<camera>_l_<light>.png`
103
+ * **JSON schema**
104
+ ```jsonc
105
+ {
106
+ "camera_angle_x": 0.3558,
107
+ "frames": [{
108
+ "file_paths": ["resized/theta_10.0_phi_0.0_board_1", …],
109
+ "light_positions": [[x,y,z], …], // metres, stage origin
110
+ "transform_matrix": [[...], ...], // 4×4 extrinsic
111
+ "width": 800, "height": 650, "cx": 400.0, "cy": 324.5
112
+ }]
113
+ }
114
+ ```
115
+ For synthetic files: identical structure, naming `r_<cam>_l_<light>`.
116
+
117
+ ## Licensing & Third-Party Assets
118
+ | Asset | Source | License / Note |
119
+ |-------|--------|----------------|
120
+ | Synthetic models | [Stanford 3-D Scans](https://graphics.stanford.edu/data/3Dscanrep/) | Varies (non-commercial / research) |
121
+ | | [BlenderKit](https://www.blenderkit.com/) | CC-0, CC-BY or Royalty-Free (check per-asset page) |
122
+ | HDR env-maps | [Poly Haven](https://polyhaven.com/) | CC-0 |
123
+ | Code | MIT (see repo) |
124
+
125
+ The dataset is released **for non-commercial research and educational use**.
126
+ If you plan to redistribute or use individual synthetic assets commercially, verify the upstream license first.
127
+
128
+ ## Citation
129
+ If you use this dataset, please cite the paper:
130
+
131
+ ```bibtex
132
+ @inproceeding{sss_gs,
133
+ author = {Dihlmann, Jan-Niklas and Majumdar, Arjun and Engelhardt, Andreas and Braun, Raphael and Lensch, Hendrik P.A.},
134
+ booktitle = {Advances in Neural Information Processing Systems},
135
+ editor = {A. Globerson and L. Mackey and D. Belgrave and A. Fan and U. Paquet and J. Tomczak and C. Zhang},
136
+ pages = {121765--121789},
137
+ publisher = {Curran Associates, Inc.},
138
+ title = {Subsurface Scattering for Gaussian Splatting},
139
+ url = {https://proceedings.neurips.cc/paper_files/paper/2024/file/dc72529d604962a86b7730806b6113fa-Paper-Conference.pdf},
140
+ volume = {37},
141
+ year = {2024}
142
+ }
143
+
144
+ ```
145
+
146
+ ## Contact & Acknowledgements
147
+ Questions, raw-capture requests, or pull-requests?
148
+ 📧 `jan-niklas.dihlmann (at) uni-tuebingen.de`
149
+
150
+ This work was funded by DFG (EXC 2064/1, SFB 1233) and the Tübingen AI Center.
other/dataset.png ADDED

Git LFS Details

  • SHA256: b4e4e89148907df8c07f7282d49d0a651929b04ebc370a5c6c6aad2b4dfe89f7
  • Pointer size: 132 Bytes
  • Size of remote file: 1.48 MB
other/preprocessing.png ADDED

Git LFS Details

  • SHA256: 0625cffadc8b1c195073c14924ff14a4185f91d68a10dca7e02dab1c1e7dcbec
  • Pointer size: 131 Bytes
  • Size of remote file: 724 kB