insailab commited on
Commit
480daad
·
verified ·
1 Parent(s): ccd0ffc

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +207 -0
README.md ADDED
@@ -0,0 +1,207 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-sa-4.0
3
+ pretty_name: Vision-Language Accessible Navigation (VLAN) Dataset
4
+ size_categories:
5
+ - 100K<n<1M
6
+ task_categories:
7
+ - other
8
+ ---
9
+
10
+ # Vision-Language Accessible Navigation (VLAN) Dataset
11
+
12
+ ## Overview
13
+
14
+ VLAN is a user-conditioned accessible navigation dataset for multimodal navigation research.
15
+ This release contains the **evaluation split** used for benchmarking navigation outputs across heterogeneous users:
16
+
17
+ - `humanoid`
18
+ - `blind` / `BVI`
19
+ - `users with mobility impairments / wheelchair users`
20
+ - `drone`
21
+ - `quadrupedRobot`
22
+
23
+ Each trajectory is paired with:
24
+
25
+ - egocentric keyframes
26
+ - a trajectory description
27
+ - user-conditioned accessibility labels
28
+ - user-specific target outputs
29
+
30
+ This dataset is designed to support two related tasks:
31
+
32
+ 1. **Action Prediction Navigation**
33
+ - `drone`
34
+ - `quadrupedRobot`
35
+ 2. **Language-Guided Navigation**
36
+ - `humanoid`
37
+ - `blind` / `BVI`
38
+ - `users with mobility impairments / wheelchair users`
39
+
40
+ ## Release Contents
41
+
42
+ This Hugging Face dataset repo is organized into two main folders:
43
+
44
+ - `vlan/`: compressed full evaluation split archives
45
+ - `samples/`: a small browsable subset of scenes and trajectories
46
+
47
+ ### `vlan/`
48
+
49
+ The `vlan/` folder contains the full evaluation split as **10 compressed archives**:
50
+
51
+ - `vlan_part01.tar.zst`
52
+ - `vlan_part02.tar.zst`
53
+ - `vlan_part03.tar.zst`
54
+ - `vlan_part04.tar.zst`
55
+ - `vlan_part05.tar.zst`
56
+ - `vlan_part06.tar.zst`
57
+ - `vlan_part07.tar.zst`
58
+ - `vlan_part08.tar.zst`
59
+ - `vlan_part09.tar.zst`
60
+ - `vlan_part10.tar.zst`
61
+
62
+ These 10 archives together contain:
63
+
64
+ Each archive can be downloaded and extracted independently.
65
+
66
+ Example extraction:
67
+
68
+ ```bash
69
+ tar --zstd -xf vlan_part01.tar.zst
70
+ ```
71
+
72
+ ### `samples/`
73
+
74
+ The `samples/` folder contains a small subset of extracted scenes for quick inspection without downloading the full benchmark.
75
+
76
+ ## Scene and Trajectory Structure
77
+
78
+ After extraction, the dataset is organized as:
79
+
80
+ ```text
81
+ <scene_name>/
82
+ <traj_name>/
83
+ KeyFrames/
84
+ 00074.png
85
+ 00089.png
86
+ ...
87
+ metadata.json
88
+ trajectory.json
89
+ target.json
90
+ ```
91
+
92
+ For example:
93
+
94
+ ```text
95
+ 0001/
96
+ 01/
97
+ KeyFrames/
98
+ 00074.png
99
+ 00089.png
100
+ ...
101
+ metadata.json
102
+ trajectory.json
103
+ target.json
104
+ ```
105
+
106
+ ## File Descriptions
107
+
108
+ ### `KeyFrames/`
109
+
110
+ Contains egocentric RGB keyframes for the trajectory.
111
+
112
+ ### `metadata.json`
113
+
114
+ Contains:
115
+
116
+ - `scene_name`
117
+ - `traj_name`
118
+ - scene taxonomy metadata:
119
+ - `lighting`
120
+ - `scene_type`
121
+ - `path_geometry`
122
+ - `weather`
123
+ - embedded keyframe metadata
124
+
125
+ ### `trajectory.json`
126
+
127
+ Contains the trajectory-level motion and obstacle information, including:
128
+
129
+ - obstacle geometry
130
+ - poses and trajectory states
131
+ - user-conditioned accessibility labels
132
+
133
+ The trajectory-level accessibility labels are stored under:
134
+
135
+ - `obstacles[0].userAccessibility`
136
+
137
+ User keys:
138
+
139
+ - `humanoid`
140
+ - `bvi`
141
+ - `wheelchair`
142
+ - `drone`
143
+ - `quadrupedRobot`
144
+
145
+ ### `target.json`
146
+
147
+ Contains the benchmark target definition for this trajectory, including:
148
+
149
+ - corrected user-conditioned accessibility labels
150
+ - user-specific ground-truth outputs
151
+ - obstacle summary
152
+ - goal information
153
+ - action references for action-prediction users
154
+
155
+ User keys in `target.json`:
156
+
157
+ - `humanoid`
158
+ - `blind`
159
+ - `wheelchair`
160
+ - `drone`
161
+ - `quadrupedRobot`
162
+
163
+
164
+ ## Accessibility Labels
165
+
166
+ The dataset uses four accessibility labels:
167
+
168
+ - `Passable`
169
+ - `Bypass`
170
+ - `HighRisk`
171
+ - `Blocked`
172
+
173
+ These labels are **user-conditioned**. The same scene and obstacle may have different labels for different users.
174
+
175
+ Examples:
176
+
177
+ - a low obstacle may be `Passable` for a drone
178
+ - the same obstacle may be `Bypass` for a humanoid user
179
+ - the same obstacle may be `HighRisk` for a blind user
180
+ - the same obstacle may be `Blocked` for a wheelchair user
181
+
182
+ ## Scene Taxonomy
183
+
184
+ Each scene is annotated with scene-level taxonomy metadata:
185
+
186
+ - `lighting`
187
+ - `scene_type`
188
+ - `path_geometry`
189
+ - `weather`
190
+
191
+ These fields are stored in each trajectory's `metadata.json`.
192
+
193
+ ## Intended Use
194
+
195
+ This dataset is intended for:
196
+
197
+ - multimodal accessible navigation
198
+ - user-conditioned navigation reasoning
199
+ - action prediction for embodied navigation agents
200
+ - language-guided instruction navigation
201
+ - accessibility-aware navigation evaluation
202
+
203
+ ## Notes
204
+
205
+ - The benchmark targets are stored in `target.json`.
206
+ - The included `samples/` folder is for quick inspection and is not a separate split.
207
+