Dataset Viewer

Split (1)

train

The dataset viewer is not available for this split.

Cannot load the dataset split (in streaming mode) to extract the first rows.

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
sequence_index: int64
random_seed: null
map_id_requested: string
map_id_loaded: string
fallback_used: bool
time_of_day: string
weather_type: string
weather_params: struct<cloudiness: double, precipitation: double, precipitation_deposits: double, wetness: double, w (... 256 chars omitted)
  child 0, cloudiness: double
  child 1, precipitation: double
  child 2, precipitation_deposits: double
  child 3, wetness: double
  child 4, wind_intensity: double
  child 5, fog_density: double
  child 6, fog_distance: double
  child 7, fog_falloff: double
  child 8, scattering_intensity: double
  child 9, mie_scattering_scale: double
  child 10, rayleigh_scattering_scale: double
  child 11, dust_storm: double
  child 12, sun_altitude_angle: double
  child 13, sun_azimuth_angle: double
fixed_delta_seconds: double
ego_spawn_retries: int64
ego_spawn_phase: int64
single_map_mode: bool
205: struct<ego_pose_matrix: list<item: double>, walker_root_pose: struct<location: struct<x: double, y:  (... 491 chars omitted)
  child 0, ego_pose_matrix: list<item: double>
      child 0, item: double
  child 1, walker_root_pose: struct<location: struct<x: double, y: double, z: double>, rotation: struct<pitch: double, yaw: doubl (... 45 chars omitted)
      child 0, location: struct<x: double, y: double, z: double>
          child 0, x: double
          child 1, y: double
          child 2, z: double
      child 1, rotation: struct<pitch: double, yaw: double, roll: double>
          child 0, pitch: double
  
...
ct<pitch: double, yaw: doubl (... 45 chars omitted)
      child 0, location: struct<x: double, y: double, z: double>
          child 0, x: double
          child 1, y: double
          child 2, z: double
      child 1, rotation: struct<pitch: double, yaw: double, roll: double>
          child 0, pitch: double
          child 1, yaw: double
          child 2, roll: double
      child 2, matrix: list<item: double>
          child 0, item: double
  child 2, camera_gait_bob: struct<enabled: bool, z_offset_m: double, amplitude_m: double, frequency_hz: double, phase_rad: doub (... 135 chars omitted)
      child 0, enabled: bool
      child 1, z_offset_m: double
      child 2, amplitude_m: double
      child 3, frequency_hz: double
      child 4, phase_rad: double
      child 5, pitch_deg: double
      child 6, roll_deg: double
      child 7, speed_mps: double
      child 8, freq_min_max_hz: list<item: double>
          child 0, item: double
      child 9, speed_min_max_mps: list<item: double>
          child 0, item: double
  child 3, ego_pose: struct<location: struct<x: double, y: double, z: double>, rotation: struct<pitch: double, yaw: doubl (... 17 chars omitted)
      child 0, location: struct<x: double, y: double, z: double>
          child 0, x: double
          child 1, y: double
          child 2, z: double
      child 1, rotation: struct<pitch: double, yaw: double, roll: double>
          child 0, pitch: double
          child 1, yaw: double
          child 2, roll: double
to
{'179': {'ego_pose_matrix': List(Value('float64')), 'walker_root_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}}}, '180': {'ego_pose_matrix': List(Value('float64')), 'walker_root_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': 
...
otation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}}}, '228': {'ego_pose_matrix': List(Value('float64')), 'walker_root_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}}}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              sequence_index: int64
              random_seed: null
              map_id_requested: string
              map_id_loaded: string
              fallback_used: bool
              time_of_day: string
              weather_type: string
              weather_params: struct<cloudiness: double, precipitation: double, precipitation_deposits: double, wetness: double, w (... 256 chars omitted)
                child 0, cloudiness: double
                child 1, precipitation: double
                child 2, precipitation_deposits: double
                child 3, wetness: double
                child 4, wind_intensity: double
                child 5, fog_density: double
                child 6, fog_distance: double
                child 7, fog_falloff: double
                child 8, scattering_intensity: double
                child 9, mie_scattering_scale: double
                child 10, rayleigh_scattering_scale: double
                child 11, dust_storm: double
                child 12, sun_altitude_angle: double
                child 13, sun_azimuth_angle: double
              fixed_delta_seconds: double
              ego_spawn_retries: int64
              ego_spawn_phase: int64
              single_map_mode: bool
              205: struct<ego_pose_matrix: list<item: double>, walker_root_pose: struct<location: struct<x: double, y:  (... 491 chars omitted)
                child 0, ego_pose_matrix: list<item: double>
                    child 0, item: double
                child 1, walker_root_pose: struct<location: struct<x: double, y: double, z: double>, rotation: struct<pitch: double, yaw: doubl (... 45 chars omitted)
                    child 0, location: struct<x: double, y: double, z: double>
                        child 0, x: double
                        child 1, y: double
                        child 2, z: double
                    child 1, rotation: struct<pitch: double, yaw: double, roll: double>
                        child 0, pitch: double
                
              ...
              ct<pitch: double, yaw: doubl (... 45 chars omitted)
                    child 0, location: struct<x: double, y: double, z: double>
                        child 0, x: double
                        child 1, y: double
                        child 2, z: double
                    child 1, rotation: struct<pitch: double, yaw: double, roll: double>
                        child 0, pitch: double
                        child 1, yaw: double
                        child 2, roll: double
                    child 2, matrix: list<item: double>
                        child 0, item: double
                child 2, camera_gait_bob: struct<enabled: bool, z_offset_m: double, amplitude_m: double, frequency_hz: double, phase_rad: doub (... 135 chars omitted)
                    child 0, enabled: bool
                    child 1, z_offset_m: double
                    child 2, amplitude_m: double
                    child 3, frequency_hz: double
                    child 4, phase_rad: double
                    child 5, pitch_deg: double
                    child 6, roll_deg: double
                    child 7, speed_mps: double
                    child 8, freq_min_max_hz: list<item: double>
                        child 0, item: double
                    child 9, speed_min_max_mps: list<item: double>
                        child 0, item: double
                child 3, ego_pose: struct<location: struct<x: double, y: double, z: double>, rotation: struct<pitch: double, yaw: doubl (... 17 chars omitted)
                    child 0, location: struct<x: double, y: double, z: double>
                        child 0, x: double
                        child 1, y: double
                        child 2, z: double
                    child 1, rotation: struct<pitch: double, yaw: double, roll: double>
                        child 0, pitch: double
                        child 1, yaw: double
                        child 2, roll: double
              to
              {'179': {'ego_pose_matrix': List(Value('float64')), 'walker_root_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}}}, '180': {'ego_pose_matrix': List(Value('float64')), 'walker_root_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': 
              ...
              otation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}}}, '228': {'ego_pose_matrix': List(Value('float64')), 'walker_root_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}, 'matrix': List(Value('float64'))}, 'camera_gait_bob': {'enabled': Value('bool'), 'z_offset_m': Value('float64'), 'amplitude_m': Value('float64'), 'frequency_hz': Value('float64'), 'phase_rad': Value('float64'), 'pitch_deg': Value('float64'), 'roll_deg': Value('float64'), 'speed_mps': Value('float64'), 'freq_min_max_hz': List(Value('float64')), 'speed_min_max_mps': List(Value('float64'))}, 'ego_pose': {'location': {'x': Value('float64'), 'y': Value('float64'), 'z': Value('float64')}, 'rotation': {'pitch': Value('float64'), 'yaw': Value('float64'), 'roll': Value('float64')}}}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Human360Occ (H3O, CARLA-generated 360° Human-Ego Occupancy Voxel Dataset)

Switch language: 中文 / Chinese README

Human360Occ is a CARLA-based 360° human-ego panoramic dataset with simulated human gait across diverse maps, weather, and times of day. It provides RGB, Z-Depth (meters), and semantic occupancy voxels (multi-resolution) per frame, plus per-frame camera pose and sequence-level metadata.
This README covers directory layout, camera & coordinate conventions, depth/semantic formats & loading, occupancy space & resolutions, and the contents of ground_truth.json and sequence_meta.json.

Dataset Snapshot

#Sequences: 160
#Frames: 8000

Time of day (time_of_day)

day: 91
night: 38
dusk: 31

Weather (weather_type)

Clear: 97
Fog: 35
Rain: 28

Maps (map_id)
Town01, Town01_Opt, Town06_Opt, Town07, Town07_Opt, Town10HD, Town10HD_Opt, Town15, Town02, Town02_Opt, Town03, Town04, Town04_Opt, Town05, Town05_Opt, Town06 (10 sequences each)

Occupancy grid resolutions

native: (64, 64, 8) — 160 sequences
fixed: (128, 128, 16) — 160 sequences

Dataset Splits

Two sequence-level splits are provided; both ensure no sequence leakage between train/val. Maps are grouped by map_id_loaded in each sequence_meta.json.

Homogeneous / within-city (homo)
- Rule: Within each map, sort sequences (priority: sequence_index, then trailing number in seq_###, then lexicographic). Take the first 8 as train, the last 2 as val.
- Ratio: With 10 sequences per map, that’s 8:2 per map, aggregating to ≈4:1 overall.
- File: splits_homo.json
  - train: concatenation of “first 8” from every map
  - val: concatenation of “last 2” from every map
  - meta.maps: per-map train/val breakdown
Heterogeneous / cross-city (heter)
- Rule: Sort map names; by default, the last 4 maps (all their sequences) form val, and the other 12 maps (all sequences) form train. You may explicitly choose the 4 val maps when generating the split.
- Note: With 16 maps × 10 sequences each, default setting is 12/4 maps (≈3:1 by sequences). If you need an exact 4:1 ratio, adjust to 13/3 maps or rebalance on the training side.
- File: splits_heter.json
  - meta.train_maps / meta.val_maps: disjoint map lists
  - train / val: aggregated sequence names across those maps

Semantic Voxel Distribution (native, 64×64×8)

Class ID	Class Name	Color (R,G,B)	Count (incl. empty)	Freq (incl. empty)	Count (excl. empty)	Freq (excl. empty)
0	Unlabeled	(0,0,0)	233727629	0.8916001472473144	0	0.0
1	Roads	(128,64,128)	8969521	0.03421600723266602	8969521	0.3156462519439938
2	SideWalks	(244,35,232)	6957890	0.026542243957519532	6957890	0.24485498165828423
3	Building	(70,70,70)	1947338	0.007428504943847656	1947338	0.06852873648081241
4	Wall	(102,102,156)	216398	0.0008254928588867187	216398	0.007615258120046363
5	Fence	(190,153,153)	488358	0.0018629379272460938	488358	0.017185797581260463
6	Pole	(153,153,153)	124025	0.00047311782836914065	124025	0.004364561540951165
7	TrafficLight	(250,170,30)	627	0.000002391815185546875	627	0.0000220647457059172
8	TrafficSign	(220,220,0)	7955	0.000030345916748046875	7955	0.00027994426170745027
9	Vegetation	(107,142,35)	2407531	0.009184001922607422	2407531	0.08472338005440597
10	Terrain	(152,251,152)	3961691	0.015112651824951171	3961691	0.13941579661949094
11	Sky	(70,130,180)	0	0.0	0	0.0
12	Pedestrian	(220,20,60)	247198	0.0009429855346679688	247198	0.008699140365249313
13	Rider	(255,0,0)	11030	0.00004207611083984375	11030	0.0003881565313178097
14	Car	(0,0,142)	775243	0.002957317352294922	775243	0.027281562448632165
15	Truck	(0,0,70)	261796	0.0009986724853515625	261796	0.009212858320297126
16	Bus	(0,60,100)	65900	0.0002513885498046875	65900	0.002319085712950468
17	Train	(0,80,100)	0	0.0	0	0.0
18	Motorcycle	(0,0,230)	14789	0.000056415557861328124	14789	0.0005204394326073516
19	Bicycle	(119,11,32)	2557	0.000009754180908203125	2557	0.00008998334094103712
20	Static	(110,190,160)	253334	0.0009663925170898438	253334	0.008915072230722213
21	Dynamic	(170,120,50)	124981	0.0004767646789550781	124981	0.004398204119730841
22	Other	(55,90,80)	976304	0.00372430419921875	976304	0.03435709647794224
23	Water	(45,60,150)	0	0.0	0	0.0
24	RoadLine	(157,234,50)	56830	0.00021678924560546874	56830	0.0019999035063273914
25	Ground	(81,0,81)	515729	0.001967350006103516	515729	0.01814900994922962
26	Bridge	(150,100,100)	4795	0.000018291473388671875	4795	0.00016874075862818654
27	RailTrack	(230,150,140)	0	0.0	0	0.0
28	GuardRail	(180,165,180)	24551	0.00009365463256835937	24551	0.0008639737987655074

Semantic Voxel Distribution (fixed128, 128×128×16)

Class ID	Class Name	Color (R,G,B)	Count (incl. empty)	Freq (incl. empty)	Count (excl. empty)	Freq (excl. empty)
0	Unlabeled	(0,0,0)	2013800865	0.960255081653595	0	0.0
1	Roads	(128,64,128)	26179410	0.012483315467834473	26179410	0.31408582498606646
2	SideWalks	(244,35,232)	20830403	0.009932710170745849	20830403	0.24991144991606892
3	Building	(70,70,70)	6478531	0.0030892043113708496	6478531	0.07772576822139254
4	Wall	(102,102,156)	690551	0.0003292803764343262	690551	0.008284842192010942
5	Fence	(190,153,153)	1349004	0.0006432552337646484	1349004	0.016184590647745828
6	Pole	(153,153,153)	395724	0.00018869590759277343	395724	0.004747673801922433
7	TrafficLight	(250,170,30)	11436	0.0000054531097412109375	11436	0.0001372026907611996
8	TrafficSign	(220,220,0)	33180	0.000015821456909179687	33180	0.0003980749632263556
9	Vegetation	(107,142,35)	6289421	0.0029990296363830568	6289421	0.07545693288999604
10	Terrain	(152,251,152)	11249802	0.005364323616027832	11249802	0.1349687919666601
11	Sky	(70,130,180)	0	0.0	0	0.0
12	Pedestrian	(220,20,60)	683598	0.0003259649276733398	683598	0.0082014240118026
13	Rider	(255,0,0)	34986	0.00001668262481689453	34986	0.0004197423346424737
14	Car	(0,0,142)	2162585	0.001031200885772705	2162585	0.025945477527090664
15	Truck	(0,0,70)	804476	0.0003836040496826172	804476	0.009651650214481182
16	Bus	(0,60,100)	212243	0.00010120534896850586	212243	0.00254637204400396
17	Train	(0,80,100)	0	0.0	0	0.0
18	Motorcycle	(0,0,230)	50946	0.000024292945861816405	50946	0.0006112214308779359
19	Bicycle	(119,11,32)	12609	0.000006012439727783203	12609	0.00015127568448827962
20	Static	(110,190,160)	732698	0.00034937763214111327	732698	0.008790498173780118
21	Dynamic	(170,120,50)	328561	0.00015667009353637694	328561	0.003941889933472412
22	Other	(55,90,80)	2861127	0.0013642916679382324	2861127	0.03432619123902752
23	Water	(45,60,150)	0	0.0	0	0.0
24	RoadLine	(157,234,50)	403207	0.00019226408004760742	403207	0.0048374506237977444
25	Ground	(81,0,81)	1475416	0.0007035331726074218	1475416	0.017701210667377235
26	Bridge	(150,100,100)	16353	0.000007797718048095704	16353	0.00019619408901870382
27	RailTrack	(230,150,140)	0	0.0	0	0.0
28	GuardRail	(180,165,180)	64868	0.00003093147277832031	64868	0.0007782497502883435

Directory Layout (per sequence `seq_xxx`)


occupancy_gt/                         # semantic occupancy (native, .npy)
occupancy_gt_128x128x16/              # semantic occupancy (128×128×16, .npy)
panorama_depth_m/                     # panoramic Z-Depth in meters (float32 .npy)
panorama_depth_raw/                   # panoramic Z-Depth in millimeters (uint16 .png)
panorama_rgb/                         # panoramic RGB (.png)
pcd_voxel_centers/                    # voxel-center point cloud (.pcd), aligned with occupancy_gt
pcd_voxel_centers_128x128x16/         # voxel-center point cloud for 128×128×16 (.pcd)
ground_truth.json                     # per-frame pose (see below)
sequence_meta.json                    # per-sequence metadata (see below)

Capture & Camera Setup

Cubemap (6 faces) + horizontal stitching: faces ordered as front/right/back/left/up/down with identical FOV (default 90°) and image size; equirect projection produces the panorama.
Horizontal mirroring: RGB/Depth/Semantic panoramas are mirrored horizontally for a consistent viewing convention; occupancy grids are mirrored on the Y axis to align with panoramas.
Coordinate frame: Unreal left-handed, Z-up. Poses follow this convention.

Weather & Time (per sequence)

sequence_meta.json logs:

map_id_requested, map_id_loaded (e.g., Town01)
time_of_day (e.g., day / night / dusk)
weather_type (e.g., Clear / Rain / Fog)
weather_params: cloudiness, precipitation, wetness, wind_intensity, fog_density, fog_distance, fog_falloff, scattering_intensity, mie_scattering_scale, rayleigh_scattering_scale, dust_storm, sun_altitude_angle, sun_azimuth_angle
fixed_delta_seconds and other run-time fields (ego_spawn_retries, ego_spawn_phase, single_map_mode, …)

Semantic Labels & Palette (CARLA 0–28)

Single-channel uint8 IDs with CARLA’s official palette.

Value	Tag	Color (R,G,B)	Description (abridged)
0	Unlabeled	(0,0,0)	Unlabeled or non-colliding
1	Roads	(128,64,128)	Drivable road surface
2	SideWalks	(244,35,232)	Pedestrian/cyclist ground (incl. curbs, islands)
3	Building	(70,70,70)	Buildings and attached elements
4	Wall	(102,102,156)	Standalone walls
5	Fence	(190,153,153)	Barriers and rails
6	Pole	(153,153,153)	Vertical poles (incl. with arms)
7	TrafficLight	(250,170,30)	Light boxes (no poles)
8	TrafficSign	(220,220,0)	Regulatory/informational signs (no poles)
9	Vegetation	(107,142,35)	Trees and vertical vegetation
10	Terrain	(152,251,152)	Grass/soil/sand (non-drivable)
11	Sky	(70,130,180)	Open sky
12	Pedestrian	(220,20,60)	Walking humans
13	Rider	(255,0,0)	Humans on vehicles/devices
14	Car	(0,0,142)	Cars/vans
15	Truck	(0,0,70)	Trucks
16	Bus	(0,60,100)	Buses
17	Train	(0,80,100)	Trains/trams
18	Motorcycle	(0,0,230)	Motorcycles
19	Bicycle	(119,11,32)	Bicycles
20	Static	(110,190,160)	Static props (hydrants, benches, stations, …)
21	Dynamic	(170,120,50)	Movable props/animals
22	Other	(55,90,80)	Everything else
23	Water	(45,60,150)	Water surfaces
24	RoadLine	(157,234,50)	Road markings
25	Ground	(81,0,81)	Ground-level structures not elsewhere
26	Bridge	(150,100,100)	Bridge structures
27	RailTrack	(230,150,140)	Non-drivable rail tracks
28	GuardRail	(180,165,180)	Guard rails/crash barriers

Depth Format (Z-Depth) & Loading

Primary: panorama_depth_m/*.npy is float32 meters, storing Z-depth along the forward axis of the cubemap face.
Compatible: panorama_depth_raw/*.png is uint16 millimeters; divide by 1000.0 to get meters.

Loader snippet

import os, cv2, numpy as np

def load_depth_m(seq_dir, fid):
    npy = os.path.join(seq_dir, 'panorama_depth_m', f'{fid:06d}.npy')
    if os.path.exists(npy):
        return np.load(npy).astype(np.float32)
    raw = os.path.join(seq_dir, 'panorama_depth_raw', f'{fid:06d}.png')
    if os.path.exists(raw):
        return cv2.imread(raw, cv2.IMREAD_UNCHANGED).astype(np.float32) / 1000.0
    raise FileNotFoundError

From Panoramic Z-Depth to Local Camera Point Cloud

For an equirect pixel, compute ray direction n=(nx,ny,nz) (right-handed: X forward / Y left / Z up). Let denom = max(|nx|, |ny|, |nz|) (projection factor to the cubemap face’s forward axis). If z_depth is the per-pixel Z-depth, Euclidean distance is r = z_depth / denom, and the 3D point is r * n.

Occupancy Grid (semantic voxels)

Space range (ego-centric): ±12.8 m left/right, +0.8 m up, −2.4 m down (approx.).
Voxel size: 0.2 m and 0.4 m variants.
Multi-resolution export: native occupancy_gt (e.g., 64×64×8) and fixed occupancy_gt_128x128x16, plus pcd_voxel_centers_* for visualization.
Voxel centers: for index (ix,iy,iz), center = base + (index + 0.5) * voxel_size.

`ground_truth.json` (pose) & Loading

Per frame we store:

ego_pose: location{x,y,z} + rotation{pitch,yaw,roll} (Unreal LH, Z-up)
ego_pose_matrix: 4×4 homogeneous matrix (same convention)
Optional: camera_gait_bob, walker_root_pose (include a matrix)

Recommendation: prefer ego_pose_matrix (reshape to 4×4). If missing, rebuild from ego_pose.

Loader snippet

import json, numpy as np

def load_pose_M(gt_path, fid):
    gt = json.load(open(gt_path, 'r', encoding='utf-8'))
    meta = gt[str(int(fid))]
    if 'ego_pose_matrix' in meta:
        return np.array(meta['ego_pose_matrix'], np.float32).reshape(4, 4)
    # Otherwise, reconstruct from ego_pose (omitted here).

`sequence_meta.json` (per-sequence metadata)

sequence_index
map_id_requested, map_id_loaded
time_of_day (e.g., day/night/dusk)
weather_type and weather_params (cloudiness, precipitation, wetness, wind_intensity, fog_density, fog_distance, fog_falloff, scattering_intensity, mie_scattering_scale, rayleigh_scattering_scale, dust_storm, sun_altitude_angle, sun_azimuth_angle)
fixed_delta_seconds, and other run-time fields (ego_spawn_retries, ego_spawn_phase, single_map_mode)

Acknowledgments

Generated using the CARLA simulator with custom capture and processing for panoramic RGB/Depth/Semantic and semantic occupancy.

Citation

If you use the Human360Occ dataset in your research, please cite OneOcc. Thank you!

@InProceedings{Shi_2026_CVPR,
  author    = {Shi, Hao and Wang, Ze and Guo, Shangwei and Duan, Mengfei and Wang, Song and Chen, Teng and Yang, Kailun and Wang, Lin and Wang, Kaiwei},
  title     = {{OneOcc}: Semantic Occupancy Prediction for Legged Robots with a Single Panoramic Camera},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month     = {June},
  year      = {2026},
  pages     = {14229--14240}
}

Downloads last month: 63

Total file size:

30.7 GB