Datasets:

Modalities:
Image
Text
Formats:
parquet
License:
degbo commited on
Commit
1c3a6e0
·
verified ·
1 Parent(s): 641859b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +193 -0
README.md CHANGED
@@ -81,3 +81,196 @@ configs:
81
  - split: train
82
  path: models/train-*
83
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
81
  - split: train
82
  path: models/train-*
83
  ---
84
+
85
+ # Olbedo Dataset
86
+
87
+ Multi-modal outdoor dataset for intrinsic image decomposition, containing RGB images, albedo, depth, surface normals, camera parameters, sky HDR maps, and 3D scene models.
88
+
89
+ **HuggingFace:** [GDAOSU/Olbedo](https://huggingface.co/datasets/GDAOSU/Olbedo)
90
+
91
+ ## Dataset Structure
92
+
93
+ ### Main config (default)
94
+
95
+ Three splits: `train` (5,664 samples), `train_selected` (2,487 samples), `test` (520 samples).
96
+
97
+ | Column | Type | Description |
98
+ |--------|------|-------------|
99
+ | `frame_id` | int | Unique frame identifier |
100
+ | `orig_frame` | string | Original frame number |
101
+ | `image` | Image | sRGB preview (PNG, for visualization) |
102
+ | `image_raw` | binary | Original EXR file (Linear RGB, for training) |
103
+ | `albedo` | Image | sRGB preview (PNG, for visualization) |
104
+ | `albedo_raw` | binary | Original EXR file (Linear RGB, for training) |
105
+ | `depth` | Image | Colormap preview (PNG, INFERNO colormap) |
106
+ | `depth_raw` | binary | Original EXR file (float, for training) |
107
+ | `normal` | Image | RGB preview (PNG, remapped from [-1,1] to [0,1]) |
108
+ | `normal_raw` | binary | Original EXR file (float RGB, for training) |
109
+ | `mask` | Image | Segmentation mask (PNG) |
110
+ | `camera` | string | Camera parameters (JSON string) |
111
+ | `sky_raw` | binary | Sky HDR map (.hdr), `None` if not available |
112
+ | `model` | string | GLB model filename (e.g. `scene_date_lighting.glb`) |
113
+ | `scene` | string | Scene name |
114
+ | `date` | string | Capture date (YYYYMMDD) |
115
+ | `lighting` | string | Lighting condition (sunrise/sunset/overcast) |
116
+
117
+ The `test` split has `None` for depth, normal, camera, sky, and model columns.
118
+
119
+ ### Models config
120
+
121
+ A separate config containing 27 3D scene models in GLB format.
122
+
123
+ | Column | Type | Description |
124
+ |--------|------|-------------|
125
+ | `model_name` | string | Model identifier (e.g. `osu_coe_corridors_20220720_sunrise`) |
126
+ | `scene` | string | Scene name |
127
+ | `date` | string | Capture date |
128
+ | `lighting` | string | Lighting condition |
129
+ | `model_raw` | binary | GLB file (binary) |
130
+
131
+ ## Usage
132
+
133
+ ### Load the dataset
134
+
135
+ ```python
136
+ from datasets import load_dataset
137
+
138
+ # Load a split (downloads data)
139
+ ds = load_dataset("GDAOSU/Olbedo", split="train")
140
+
141
+ # Or use streaming to avoid downloading everything
142
+ ds = load_dataset("GDAOSU/Olbedo", split="train", streaming=True)
143
+ row = next(iter(ds))
144
+ ```
145
+
146
+ ### Recover raw EXR files (image, albedo, depth, normal)
147
+
148
+ ```python
149
+ ds = load_dataset("GDAOSU/Olbedo", split="train", streaming=True)
150
+
151
+ for row in ds:
152
+ fid = row['frame_id']
153
+
154
+ # Save image EXR
155
+ with open(f"{fid:04d}_im.exr", "wb") as f:
156
+ f.write(row['image_raw'])
157
+
158
+ # Save albedo EXR
159
+ with open(f"{fid:04d}_albedo.exr", "wb") as f:
160
+ f.write(row['albedo_raw'])
161
+
162
+ # Save depth EXR
163
+ if row['depth_raw'] is not None:
164
+ with open(f"{fid:04d}_depth.exr", "wb") as f:
165
+ f.write(row['depth_raw'])
166
+
167
+ # Save normal EXR
168
+ if row['normal_raw'] is not None:
169
+ with open(f"{fid:04d}_normal.exr", "wb") as f:
170
+ f.write(row['normal_raw'])
171
+ ```
172
+
173
+ ### Recover camera JSON
174
+
175
+ ```python
176
+ import json
177
+
178
+ row = next(iter(ds))
179
+ if row['camera'] is not None:
180
+ camera = json.loads(row['camera'])
181
+ print(camera['focal'], camera['cx'], camera['cy']) # intrinsics
182
+ print(camera['X'], camera['Y'], camera['Z']) # translation
183
+
184
+ # Save to file
185
+ fid = row['frame_id']
186
+ with open(f"{fid:04d}_camera.json", "w") as f:
187
+ f.write(row['camera'])
188
+ ```
189
+
190
+ ### Recover sky HDR maps
191
+
192
+ Only some frames have sky HDR maps (frames 3801+).
193
+
194
+ ```python
195
+ for row in ds:
196
+ if row['sky_raw'] is not None:
197
+ fid = row['frame_id']
198
+ with open(f"{fid:04d}_sky.hdr", "wb") as f:
199
+ f.write(row['sky_raw'])
200
+ ```
201
+
202
+ ### Recover 3D models (GLB)
203
+
204
+ Models are stored in a separate config. Each model corresponds to a unique scene/date/lighting combination.
205
+
206
+ ```python
207
+ models = load_dataset("GDAOSU/Olbedo", "models", split="train", streaming=True)
208
+
209
+ for row in models:
210
+ name = row['model_name']
211
+ with open(f"{name}.glb", "wb") as f:
212
+ f.write(row['model_raw'])
213
+ print(f"Saved {name}.glb ({len(row['model_raw']) / 1e6:.0f} MB)")
214
+ ```
215
+
216
+ ### Link frames to their 3D model
217
+
218
+ Each frame's `model` field contains the GLB filename. To find which model a frame belongs to:
219
+
220
+ ```python
221
+ row = next(iter(ds))
222
+ print(row['model']) # e.g. "osu_coe_corridors_20220720_sunrise.glb"
223
+ ```
224
+
225
+ ### Recover all data for a single frame
226
+
227
+ ```python
228
+ import json
229
+
230
+ ds = load_dataset("GDAOSU/Olbedo", split="train", streaming=True)
231
+ row = next(iter(ds))
232
+ fid = row['frame_id']
233
+ prefix = f"{fid:04d}"
234
+
235
+ # Save all files
236
+ with open(f"{prefix}_im.exr", "wb") as f:
237
+ f.write(row['image_raw'])
238
+ with open(f"{prefix}_albedo.exr", "wb") as f:
239
+ f.write(row['albedo_raw'])
240
+ if row['depth_raw']:
241
+ with open(f"{prefix}_depth.exr", "wb") as f:
242
+ f.write(row['depth_raw'])
243
+ if row['normal_raw']:
244
+ with open(f"{prefix}_normal.exr", "wb") as f:
245
+ f.write(row['normal_raw'])
246
+ if row['camera']:
247
+ with open(f"{prefix}_camera.json", "w") as f:
248
+ f.write(row['camera'])
249
+ if row['sky_raw']:
250
+ with open(f"{prefix}_sky.hdr", "wb") as f:
251
+ f.write(row['sky_raw'])
252
+
253
+ # Save mask from preview Image
254
+ if row['mask'] is not None:
255
+ row['mask'].save(f"{prefix}_mask.png")
256
+
257
+ print(f"Frame {fid}: scene={row['scene']}, date={row['date']}, lighting={row['lighting']}, model={row['model']}")
258
+ ```
259
+
260
+ ## Scenes
261
+
262
+ The dataset covers 4 locations with multiple captures under different dates and lighting conditions:
263
+
264
+ - `goodale_park`
265
+ - `osu_coe_corridors`
266
+ - `osu_residential_area`
267
+ - `schottenstein_center`
268
+
269
+ ## File Formats
270
+
271
+ - **Image/Albedo EXR**: Linear RGB, float16/float32. Apply sRGB transfer function for display.
272
+ - **Depth EXR**: Single-channel float (channel name `I`). Use OpenEXR library to read.
273
+ - **Normal EXR**: RGB float in [-1, 1] range. Remap to [0, 1] for visualization.
274
+ - **Sky HDR**: Radiance HDR format (.hdr).
275
+ - **3D Models**: glTF Binary (.glb).
276
+ - **Camera JSON**: Contains intrinsics (focal, cx, cy, distortion), extrinsics (rotation matrix, translation), and metadata (GPS, sun position).