Datasets:

Modalities:
Image
Languages:
English
ArXiv:
Tags:
code
License:
yueyin27 commited on
Commit
321b381
·
verified ·
1 Parent(s): 8c4397e

Upload README.md

Browse files
Files changed (1) hide show
  1. README.md +74 -7
README.md CHANGED
@@ -1,7 +1,74 @@
1
- ---
2
- language:
3
- - en
4
- tags:
5
- - Novel-View-Synthesis
6
- - 3D-Reconstruction
7
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # RefRef: A Synthetic Dataset and Benchmark for Reconstructing Scenes with Refractive and Reflective Objects
2
+
3
+ ## Overview
4
+ **RefRef** is a synthetic dataset and benchmark for reconstructing scenes with **Ref**ractive-**Ref**lective objects. Our dataset includes 50 objects with varying complexity, from single-material to multi-material compositions with different Indices of Refraction (IoRs) and transparency levels. Each object is placed in two distinct scenes featuring ornate textures and patterns, resulting in 100 scenes in total.
5
+
6
+ ## Directory Structure
7
+
8
+ ```plaintext
9
+ RefRef_Dataset/
10
+ ├── README.md
11
+ ├── dataset_info/ # Metadata and dataset description files
12
+ │ ├── object_list.txt
13
+ │ ├── scene_list.txt
14
+ │ └── IoR_info.json # IoR values mapped to each object
15
+ ├── image_data/ # Rendered images, depth maps, and masks for each object
16
+ │ ├── textured_cube_scene/
17
+ │ │ └── {simple_shapes, complex_shapes, household_items, lab_equipment}/
18
+ │ │ └── {object_name}/
19
+ │ │ ├── train/ # Training set
20
+ │ │ │ ├── r_0.png # RGB image
21
+ │ │ │ ├── r_0_depth_0000.png # Depth map
22
+ │ │ │ ├── r_0_mask_0000.png # Mask
23
+ │ │ │ ├── r_1.png
24
+ │ │ │ ├── r_1_depth_0000.png
25
+ │ │ │ ├── r_1_mask_0000.png
26
+ │ │ │ └── ...
27
+ │ │ ├── val/ # Validation set
28
+ │ │ ├── test/ # Testing set
29
+ │ │ ├── transforms_train.json
30
+ │ │ ├── transforms_val.json
31
+ │ │ └── transforms_test.json
32
+ │ ├── textured_sphere_scene/
33
+ │ └── ...
34
+ ├── mesh_files/ # 3D mesh files (.ply format) for each object
35
+ │ └── {simple_shapes, complex_shapes, household_items, lab_equipment}/
36
+ │ └── ...
37
+ ├── blender_files/ # Blender source files for each object, organised by scene
38
+ │ ├── bgpanels_cube/ # Background panels for cube scene
39
+ │ ├── bgpanels_sphere/ # Background panels for sphere scene
40
+ │ └── {textured_cube_scene, textured_sphere_scene}/
41
+ │ └── ...
42
+ └── benchmarks/ # Benchmark results from various methods
43
+ ├── oracle_method/
44
+ ├── Zip-NeRF/
45
+ ├── Ray Deformation/
46
+ ├── MS-NeRF/
47
+ ├── NeUS/
48
+ └── ...
49
+ ```
50
+
51
+ ## Object and Scenes
52
+
53
+ The dataset includes 50 objects categorised into four groups based on their complexity, material composition, and shape:
54
+ 1. **simple_shapes**: Basic geometries with single material.
55
+ 2. **complex_shapes**: Intricate geometries with single material, such as sculptures.
56
+ 3. **household_items**: Objects with potentially mixed materials, including items like wine glasses, plastic bottles, and teapots.
57
+ 4. **lab_equipment**: Laboratory objects that may include single and multi-material compositions, such as beakers and test tubes.
58
+
59
+ Each object is placed in two distinct scenes:
60
+ - `textured_cube_scene/`: Objects placed within a textured cube environment.
61
+ - `textured_sphere_scene/`: Objects placed within a textured sphere environment.
62
+
63
+ ## IoR Information
64
+
65
+ A single JSON file, `IoR_info.json`, will be provided in the `dataset_info/` directory. This file maps each object (by filename) to its Index of Refraction (IoR) values. This helps in identifying the IoR for each `.ply` file in `mesh_files`.
66
+
67
+ Example format for `IoR_info.json`:
68
+ ```json
69
+ {
70
+ "cube.ply": {"IoR": 1.5},
71
+ "diamond.ply": {"IoR": 2.4},
72
+ "wine_glass_with_wine.ply": {"IoR": {"glass": 1.5, "wine": 1.33}},
73
+ ...
74
+ }