Dataset Viewer
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
evidence_id: string
measurement_scope: struct<segment_count: int64, segment_relationship: string, independent_experimental_sessions: bool>
child 0, segment_count: int64
child 1, segment_relationship: string
child 2, independent_experimental_sessions: bool
timestamp_integrity: struct<stream_checks: int64, monotonicity_violations: int64, duplicate_timestamps: int64>
child 0, stream_checks: int64
child 1, monotonicity_violations: int64
child 2, duplicate_timestamps: int64
temporal_coverage: struct<video_imu_bracket_coverage_segments: int64, evaluated_segments: int64, full_bracket_coverage: (... 169 chars omitted)
child 0, video_imu_bracket_coverage_segments: int64
child 1, evaluated_segments: int64
child 2, full_bracket_coverage: bool
child 3, imu_pre_roll_ms: struct<min: double, median: double, mean: double, max: double>
child 0, min: double
child 1, median: double
child 2, mean: double
child 3, max: double
child 4, imu_post_roll_ms: struct<min: double, median: double, mean: double, max: double>
child 0, min: double
child 1, median: double
child 2, mean: double
child 3, max: double
observed_gaps: struct<video_segments_with_max_gap_over_100ms: int64, video_max_observed_gap_ms: double, app_level_a (... 208 chars omitted)
child 0, video_segments_with_max_gap_over_100ms: int64
child 1, video_max_observed_gap_ms: double
child 2, app_level_accelerometer_segments_with_gap_over_100ms: int64
child 3, app_level_accelerometer_max_observed_gap_ms: double
child 4, app_level_gyroscope_segments_with_gap_over_100ms: int64
child 5, app_level_gyroscope_max_observed_gap_ms: double
claim_boundary: string
release_type: string
note: string
capture_context: struct<recording_structure: string, task_context: string, country: string>
child 0, recording_structure: string
child 1, task_context: string
child 2, country: string
privacy_and_provenance: struct<description: string, internal_records_retained: bool, internal_operational_metadata_reproduce (... 170 chars omitted)
child 0, description: string
child 1, internal_records_retained: bool
child 2, internal_operational_metadata_reproduced: bool
child 3, precise_gps_included: bool
child 4, contributor_identifiers_included: bool
child 5, private_operational_identifiers_included: bool
child 6, customer_specific_information_included: bool
to
{'evidence_id': Value('string'), 'release_type': Value('string'), 'capture_context': {'recording_structure': Value('string'), 'task_context': Value('string'), 'country': Value('string')}, 'privacy_and_provenance': {'description': Value('string'), 'internal_records_retained': Value('bool'), 'internal_operational_metadata_reproduced': Value('bool'), 'precise_gps_included': Value('bool'), 'contributor_identifiers_included': Value('bool'), 'private_operational_identifiers_included': Value('bool'), 'customer_specific_information_included': Value('bool')}, 'note': Value('string')}
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 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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
evidence_id: string
measurement_scope: struct<segment_count: int64, segment_relationship: string, independent_experimental_sessions: bool>
child 0, segment_count: int64
child 1, segment_relationship: string
child 2, independent_experimental_sessions: bool
timestamp_integrity: struct<stream_checks: int64, monotonicity_violations: int64, duplicate_timestamps: int64>
child 0, stream_checks: int64
child 1, monotonicity_violations: int64
child 2, duplicate_timestamps: int64
temporal_coverage: struct<video_imu_bracket_coverage_segments: int64, evaluated_segments: int64, full_bracket_coverage: (... 169 chars omitted)
child 0, video_imu_bracket_coverage_segments: int64
child 1, evaluated_segments: int64
child 2, full_bracket_coverage: bool
child 3, imu_pre_roll_ms: struct<min: double, median: double, mean: double, max: double>
child 0, min: double
child 1, median: double
child 2, mean: double
child 3, max: double
child 4, imu_post_roll_ms: struct<min: double, median: double, mean: double, max: double>
child 0, min: double
child 1, median: double
child 2, mean: double
child 3, max: double
observed_gaps: struct<video_segments_with_max_gap_over_100ms: int64, video_max_observed_gap_ms: double, app_level_a (... 208 chars omitted)
child 0, video_segments_with_max_gap_over_100ms: int64
child 1, video_max_observed_gap_ms: double
child 2, app_level_accelerometer_segments_with_gap_over_100ms: int64
child 3, app_level_accelerometer_max_observed_gap_ms: double
child 4, app_level_gyroscope_segments_with_gap_over_100ms: int64
child 5, app_level_gyroscope_max_observed_gap_ms: double
claim_boundary: string
release_type: string
note: string
capture_context: struct<recording_structure: string, task_context: string, country: string>
child 0, recording_structure: string
child 1, task_context: string
child 2, country: string
privacy_and_provenance: struct<description: string, internal_records_retained: bool, internal_operational_metadata_reproduce (... 170 chars omitted)
child 0, description: string
child 1, internal_records_retained: bool
child 2, internal_operational_metadata_reproduced: bool
child 3, precise_gps_included: bool
child 4, contributor_identifiers_included: bool
child 5, private_operational_identifiers_included: bool
child 6, customer_specific_information_included: bool
to
{'evidence_id': Value('string'), 'release_type': Value('string'), 'capture_context': {'recording_structure': Value('string'), 'task_context': Value('string'), 'country': Value('string')}, 'privacy_and_provenance': {'description': Value('string'), 'internal_records_retained': Value('bool'), 'internal_operational_metadata_reproduced': Value('bool'), 'precise_gps_included': Value('bool'), 'contributor_identifiers_included': Value('bool'), 'private_operational_identifiers_included': Value('bool'), 'customer_specific_information_included': Value('bool')}, 'note': Value('string')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Video + IMU Capture Evaluation
Origin Data Lab — Public Technical Evidence
This repository provides measured capture characteristics from 12 consecutive server-side segments generated during one continuous real-world field recording.
The 12 segments represent consecutive units from a single continuous capture and should not be interpreted as 12 independent experimental sessions.
Evidence Overview
The evaluated recording originates from a real-world agricultural produce sorting and portioning task in South Korea.
This public release focuses on measurable video and 6-DoF IMU capture characteristics, including:
- effective video frame rate
- native accelerometer and gyroscope rates
- application-level IMU rates
- observed timing gaps
- timestamp monotonicity
- duplicate timestamp checks
- video/IMU temporal bracket coverage
Key Measured Results
| Metric | Result |
|---|---|
| Evaluated segments | 12 |
| Effective video FPS | 29.8334–30.0004 fps |
| Native accelerometer rate | 409.6461–410.1447 Hz |
| Native gyroscope rate | 409.9429–410.2659 Hz |
| App-level accelerometer rate | 97.7867–100.7711 Hz |
| App-level gyroscope rate | 97.7937–100.6345 Hz |
| Timestamp monotonicity violations | 0 |
| Duplicate timestamps | 0 |
| Video/IMU temporal bracket coverage | 12 / 12 segments |
The native IMU rates above are measured observations for this evaluated configuration, not a guaranteed hardware specification.
Evidence Artifacts
This dataset repository includes:
measurement_summary.csv— aggregate measured statistics across the 12 evaluated segmentsmeasurement_methodology.md— measurement definitions and calculation methodologytimestamp_integrity_report.json— timestamp integrity, temporal coverage, and observed gap summarypublic_capture_context.json— sanitized public description of the capture contexttechnical_boundaries.md— explicit technical limitations and claim boundaries
Aggregate Measurement Results
| Metric | Min | Median | Mean | Max |
|---|---|---|---|---|
| Effective video FPS | 29.8334 | 29.9837 | 29.9646 | 30.0004 |
| Video frame max gap (ms) | 66.63 | 66.65 | 102.74 | 499.80 |
| Native accel (Hz) | 409.6461 | 409.8665 | 409.8693 | 410.1447 |
| Native gyro (Hz) | 409.9429 | 410.1112 | 410.1097 | 410.2659 |
| Native accel max gap (ms) | 25.98 | 28.02 | 27.50 | 30.46 |
| Native gyro max gap (ms) | 4.93 | 7.34 | 10.94 | 24.68 |
| App-level accel (Hz) | 97.7867 | 99.0855 | 99.1258 | 100.7711 |
| App-level gyro (Hz) | 97.7937 | 99.2673 | 99.1824 | 100.6345 |
| App-level accel max gap (ms) | 163.80 | 205.78 | 280.85 | 589.34 |
| App-level gyro max gap (ms) | 103.65 | 207.00 | 276.75 | 599.08 |
| IMU pre-roll (ms) | 303.17 | 422.30 | 429.58 | 562.22 |
| IMU post-roll (ms) | 87.40 | 114.37 | 111.53 | 135.30 |
Timestamp Integrity
Across the evaluated streams:
- 48 timestamp stream checks
- 0 monotonicity violations
- 0 duplicate timestamps
- 12 / 12 segments showed native IMU temporal bracket coverage around the evaluated video interval
Temporal bracket coverage indicates capture-session timing coverage. It does not establish hardware frame-level video-IMU synchronization.
Observed Timing Limitations
One of the 12 evaluated segments showed a video inter-frame gap above 100 ms.
- Maximum observed video gap: 499.80 ms
Application-level IMU streams showed gaps above 100 ms in all 12 evaluated segments.
- Maximum app-level accelerometer gap: 589.34 ms
- Maximum app-level gyroscope gap: 599.08 ms
The cause of these application-level timing gaps has not been established. This public evaluation does not attribute them to buffering, batching, Android scheduling, or another specific mechanism.
Technical Boundaries
This evaluation does not claim:
- hardware frame-level video-IMU synchronization
- sub-frame synchronization
- camera-exposure-level timestamp synchronization
- a guaranteed 410 Hz hardware specification
- guaranteed continuous 100 Hz application-level delivery
- 12 independent experimental replications
- universal performance across all devices or capture configurations
See technical_boundaries.md for the detailed claim boundary.
Public Release Context
The evaluated material originates from a continuous real-world agricultural produce sorting and portioning task recorded in South Korea.
The public capture-context description is sanitized for public release and is independently described from internal operational record-keeping fields.
Precise GPS coordinates, contributor identifiers, private operational identifiers, and customer-specific information are not included in this public evidence package.
Evidence Network
Website Technical Field Note
https://origindatalab.io/technical-field-notes/video-imu-capture-evaluation.html
GitHub Technical Evidence
https://github.com/origin-data-lab/video-imu-capture-evaluation
Origin Data Lab — Real-World Data Operations
https://github.com/origin-data-lab/real-world-data-operations
About Origin Data Lab
Origin Data Lab designs and operates custom real-world data collection and dataset production programs for Physical AI, robotics, embodied AI, computer vision, and autonomous systems.
Capabilities include egocentric human demonstrations, multimodal sensor capture, IMU and video data, metadata engineering, field operations, quality control, technical validation, and structured dataset delivery.
Evidence Status: Public Technical Evaluation
Technical Field Note: #01
- Downloads last month
- 179