WalkerCH commited on
Commit
c080fe0
·
1 Parent(s): f8bd9f8

docs: add GSO30 dataset readme

Browse files
Files changed (1) hide show
  1. README.md +93 -0
README.md CHANGED
@@ -1,3 +1,96 @@
1
  ---
2
  license: mit
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: mit
3
  ---
4
+
5
+ # Streaming3D Dataset
6
+
7
+ This dataset contains assets used by the Streaming3D benchmark. The current
8
+ release documents the `GSO30` subset; other subsets may be added later.
9
+
10
+ ## GSO30
11
+
12
+ `GSO30` is a 30-object subset derived from Google Scanned Objects. Each object
13
+ directory contains training renders, evaluation assets, and the original object
14
+ mesh/material files.
15
+
16
+ ### Object List
17
+
18
+ ```text
19
+ alarm backpack bell blocks chicken cream elephant grandfather grandmother hat
20
+ leather lion lunch_bag mario oil school_bus1 school_bus2 shoe shoe1 shoe2
21
+ shoe3 soap sofa sorter sorting_board stucking_cups teapot toaster train turtle
22
+ ```
23
+
24
+ ### Directory Structure
25
+
26
+ ```text
27
+ GSO30/
28
+ <object_id>/
29
+ meshes/
30
+ model.glb
31
+ model.obj
32
+ model.mtl
33
+ texture.png
34
+ render_spiral_100/
35
+ images/
36
+ 000.png ... 099.png
37
+ masks/
38
+ 000.png ... 099.png
39
+ model/
40
+ 000.png ... 099.png
41
+ 000.npy ... 099.npy
42
+ transforms.json
43
+ model_norm.obj
44
+ model_norm.mtl
45
+ render_mvs_25/
46
+ model_norm.glb
47
+ model_norm.obj
48
+ model_norm.mtl
49
+ model/
50
+ 000.png ... 024.png
51
+ 000.npy ... 024.npy
52
+ ```
53
+
54
+ Some object folders also include auxiliary metadata, thumbnails, or legacy
55
+ render folders. The benchmark protocol uses the paths above.
56
+
57
+ ### Usage
58
+
59
+ For training or reconstruction input, use all 100 images from:
60
+
61
+ ```text
62
+ GSO30/<object_id>/render_spiral_100/images/{000..099}.png
63
+ ```
64
+
65
+ The corresponding masks are stored in:
66
+
67
+ ```text
68
+ GSO30/<object_id>/render_spiral_100/masks/{000..099}.png
69
+ ```
70
+
71
+ Camera metadata for the 100 spiral views is available in:
72
+
73
+ ```text
74
+ GSO30/<object_id>/render_spiral_100/transforms.json
75
+ GSO30/<object_id>/render_spiral_100/model/{000..099}.npy
76
+ ```
77
+
78
+ For evaluation, use the normalized GLB mesh and the 25 provided camera views
79
+ from `render_mvs_25`:
80
+
81
+ ```text
82
+ GSO30/<object_id>/render_mvs_25/model_norm.glb
83
+ GSO30/<object_id>/render_mvs_25/model/{000..024}.npy
84
+ ```
85
+
86
+ The matching reference renders for those views are:
87
+
88
+ ```text
89
+ GSO30/<object_id>/render_mvs_25/model/{000..024}.png
90
+ ```
91
+
92
+ In short, the default protocol is:
93
+
94
+ 1. Train or reconstruct from all `render_spiral_100/images` frames.
95
+ 2. Evaluate by rendering or comparing against `render_mvs_25/model_norm.glb`
96
+ using the 25 camera poses in `render_mvs_25/model/*.npy`.