hardik01shah commited on
Commit
ba4507b
·
verified ·
1 Parent(s): 4109fbc

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +209 -3
README.md CHANGED
@@ -1,3 +1,209 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ task_categories:
4
+ - image-segmentation
5
+ - object-detection
6
+ - robotics
7
+ language:
8
+ - en
9
+ tags:
10
+ - robotics
11
+ - navigation
12
+ - frontiers
13
+ - autonomous-systems
14
+ - field-robotics
15
+ - vision-foundation-models
16
+ - outdoor-navigation
17
+ - traversability
18
+ - exploration
19
+ pretty_name: WildOS Frontiers Dataset
20
+ size_categories:
21
+ - n<1K
22
+ configs:
23
+ - config_name: default
24
+ data_files:
25
+ - split: train
26
+ path: "**"
27
+ ---
28
+
29
+ # WildOS Frontiers Dataset
30
+
31
+ <div align="center">
32
+ <img src="https://leggedrobotics.github.io/wildos/static/images/Teaser-V.svg" alt="WildOS Teaser" width="800"/>
33
+ </div>
34
+
35
+ ## Dataset Description
36
+
37
+ This dataset provides **visual frontier annotations** for outdoor long-range navigation, created for [WildOS: Open-Vocabulary Object Search in the Wild](https://leggedrobotics.github.io/wildos/). The annotations are built on top of images from the [GrandTour Dataset](https://huggingface.co/datasets/leggedrobotics/grand_tour_dataset).
38
+
39
+ **Visual Frontiers** denote regions in the image that correspond to candidate locations for further exploration — such as the end of a trail, an opening
40
+ between trees, or a road turning at a curve. This dataset enables training of models to predict visual frontiers from RGB images, extending navigation reasoning beyond the geometric depth horizon.
41
+
42
+ ## Dataset Structure
43
+
44
+ ```
45
+ wildos/
46
+ ├── annotations/ # Frontier annotations (362 JSON files)
47
+ │ └── annotation_00000.json ... annotation_00389.json
48
+ ├── RGB_frames/ # Raw RGB frames (390 images + metadata)
49
+ │ ├── metadata.json # Maps to original GrandTour images
50
+ │ └── rgb_00000.png ... rgb_00389.png
51
+ ├── RGB_rectified/ # Rectified RGB images (390 images)
52
+ │ └── rect_00000.png ... rect_00389.png
53
+ └── SAM_boundaries/ # SAM-2 boundary masks (390 images)
54
+ └── bound_00000.png ... bound_00389.png
55
+ ```
56
+
57
+ ### File Descriptions
58
+
59
+ | Folder | Description | Count |
60
+ |--------|-------------|-------|
61
+ | `annotations/` | JSON files containing frontier bounding box annotations | 362 |
62
+ | `RGB_frames/` | Original RGB frames from GrandTour dataset | 390 + 1 metadata |
63
+ | `RGB_rectified/` | Rectified (undistorted) RGB images | 390 |
64
+ | `SAM_boundaries/` | Binary masks from SAM-2 boundary detection | 390 |
65
+
66
+ > **Note:** Some images do not have corresponding annotations (362 out of 390 images are annotated). Images without annotations were excluded during quality control.
67
+
68
+ ## Annotation Format
69
+
70
+ Each annotation file contains a list of frontier detections with the following structure:
71
+
72
+ ```json
73
+ [
74
+ {
75
+ "label": "frontier",
76
+ "start": [1326.0, 618.0],
77
+ "end": [1352.0, 636.0]
78
+ }
79
+ ]
80
+ ```
81
+
82
+ | Field | Description |
83
+ |-------|-------------|
84
+ | `label` | Frontier type (`"frontier"` or `"image_boundary_frontier"`) |
85
+ | `start` | Top-left corner `[x, y]` of the bounding box |
86
+ | `end` | Bottom-right corner `[x, y]` of the bounding box |
87
+
88
+ ## Usage
89
+
90
+ ### Loading the Dataset
91
+
92
+ ```python
93
+ from datasets import load_dataset
94
+
95
+ # Load the full dataset
96
+ dataset = load_dataset("leggedrobotics/wildos")
97
+
98
+ # Access specific splits
99
+ print(dataset)
100
+ ```
101
+
102
+ ### Loading Individual Files
103
+
104
+ ```python
105
+ import json
106
+ from PIL import Image
107
+
108
+ # Load an annotation
109
+ with open("wildos/annotations/annotation_00000.json", "r") as f:
110
+ annotations = json.load(f)
111
+
112
+ # Load corresponding image
113
+ image = Image.open("wildos/RGB_rectified/rect_00000.png")
114
+
115
+ print(f"Image size: {image.size}")
116
+ print(f"Number of frontiers: {len(annotations)}")
117
+ ```
118
+
119
+ ### Visualizing Annotations
120
+
121
+ We provide a visualization script to overlay frontier annotations on images:
122
+
123
+ ```python
124
+ import os
125
+ import json
126
+ import cv2
127
+ import numpy as np
128
+
129
+ def visualize_frontiers(image_path, annotation_path, output_path=None):
130
+ """Draw frontier annotations on an image."""
131
+ # Load image
132
+ img = cv2.imread(image_path)
133
+
134
+ # Load annotations
135
+ with open(annotation_path, "r") as f:
136
+ annotations = json.load(f)
137
+
138
+ # Draw each frontier
139
+ for ann in annotations:
140
+ x1, y1 = int(ann["start"][0]), int(ann["start"][1])
141
+ x2, y2 = int(ann["end"][0]), int(ann["end"][1])
142
+ color = (0, 0, 255) # Red in BGR
143
+
144
+ # Draw semi-transparent rectangle
145
+ overlay = img.copy()
146
+ cv2.rectangle(overlay, (x1, y1), (x2, y2), color, -1)
147
+ cv2.addWeighted(overlay, 0.35, img, 0.65, 0, img)
148
+ cv2.rectangle(img, (x1, y1), (x2, y2), color, 2)
149
+
150
+ if output_path:
151
+ cv2.imwrite(output_path, img)
152
+
153
+ return img
154
+
155
+ # Example usage
156
+ visualize_frontiers(
157
+ "wildos/RGB_rectified/rect_00000.png",
158
+ "wildos/annotations/annotation_00000.json",
159
+ "output_visualization.png"
160
+ )
161
+ ```
162
+
163
+ ### Metadata Mapping
164
+
165
+ The `metadata.json` file in `RGB_frames/` maps each image to its source in the GrandTour dataset:
166
+
167
+ ```python
168
+ import json
169
+
170
+ with open("wildos/RGB_frames/metadata.json", "r") as f:
171
+ metadata = json.load(f)
172
+
173
+ # Find original GrandTour image for a specific frame
174
+ original_path = metadata["rgb_00000.png"]
175
+ print(f"Original GrandTour path: {original_path}")
176
+ ```
177
+
178
+ ## Dataset Statistics
179
+
180
+ - **Total Images**: 390
181
+ - **Total Annotations**: 362 JSON files
182
+ - **Frontier Types**: `frontier`, `image_boundary_frontier`
183
+ - **Image Format**: PNG
184
+ - **Source Dataset**: [GrandTour Dataset](https://huggingface.co/datasets/leggedrobotics/grand_tour_dataset)
185
+
186
+ ## Related Resources
187
+
188
+ - **Project Page**: [WildOS: Open-Vocabulary Object Search in the Wild](https://leggedrobotics.github.io/wildos/)
189
+ - **Source Dataset**: [GrandTour Dataset](https://huggingface.co/datasets/leggedrobotics/grand_tour_dataset)
190
+
191
+ ## Citation
192
+
193
+ If you use this dataset in your research, please cite:
194
+
195
+ ```bibtex
196
+ @misc{shah2026wildosopenvocabularyobjectsearch,
197
+ title={WildOS: Open-Vocabulary Object Search in the Wild},
198
+ author={Hardik Shah and Erica Tevere and Deegan Atha and Marcel Kaufmann and Shehryar Khattak and Manthan Patel and Marco Hutter and Jonas Frey and Patrick Spieler},
199
+ year={2026},
200
+ eprint={2602.19308},
201
+ archivePrefix={arXiv},
202
+ primaryClass={cs.RO},
203
+ url={https://arxiv.org/abs/2602.19308},
204
+ }
205
+ ```
206
+
207
+ ## License
208
+
209
+ This dataset is released under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0).