kaimingyang/DVD_for_Dream.exe
Robotics • Updated • 1
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
action_path: string
config_report: struct<dtw_penalty: string, exec_sr_pos_tol: double, exec_sr_rot_tol: double, executed_smoothness_pa (... 244 chars omitted)
child 0, dtw_penalty: string
child 1, exec_sr_pos_tol: double
child 2, exec_sr_rot_tol: double
child 3, executed_smoothness_path_length_normalized: bool
child 4, executed_smoothness_resample_points: int64
child 5, pos_tol: double
child 6, rot_tol: double
child 7, smoothing: struct<enabled: bool, method: null, params: null>
child 0, enabled: bool
child 1, method: null
child 2, params: null
child 8, tracking_ndtw_max_points: int64
child 9, tracking_ndtw_orientation_fallback: string
correction: struct<correction_exhausted_count: int64, correction_exhausted_rate: double, correction_steps: struc (... 103 chars omitted)
child 0, correction_exhausted_count: int64
child 1, correction_exhausted_rate: double
child 2, correction_steps: struct<count: int64, max: double, mean: double, median: double, p95: double>
child 0, count: int64
child 1, max: double
child 2, mean: double
child 3, median: double
child 4, p95: double
child 3, total_correction_steps: int64
cosmos_policy: struct<checkpoint_match_policy: string, control_hz: double, exec_sr_pos_tol: double, exec_sr_rot_tol (... 271 chars omitted)
child 0, checkpoint_match_policy: string
child 1, control_hz: double
child 2, exec_sr_pos_tol: double
child 3, exec_sr_rot_tol: double
child 4, ndtw_max_
...
uses_controller_scaling: bool
child 1, checkpoint_match_diagnostics: list<item: struct<checkpoint_index: int64, match_mode: string, position_error_m: double, rotation_er (... 32 chars omitted)
child 0, item: struct<checkpoint_index: int64, match_mode: string, position_error_m: double, rotation_error_rad: do (... 20 chars omitted)
child 0, checkpoint_index: int64
child 1, match_mode: string
child 2, position_error_m: double
child 3, rotation_error_rad: double
child 4, success: bool
child 2, control_hz: double
child 3, errors: list<item: null>
child 0, item: null
child 4, executed_count: int64
child 5, executed_dense_count_for_ndtw: int64
child 6, executed_dense_count_full: int64
child 7, ok: bool
child 8, planned_count: int64
child 9, planned_dense_count_for_ndtw: int64
child 10, planned_dense_count_full: int64
child 11, policy_hz: double
child 12, sim_hz: double
child 13, tcp_site_name: string
child 14, warnings: list<item: null>
child 0, item: null
sanitization: struct<dropped_invalid_initial_targets: int64, executed_count: int64, initial_target_to_exec0_dist_m (... 110 chars omitted)
child 0, dropped_invalid_initial_targets: int64
child 1, executed_count: int64
child 2, initial_target_to_exec0_dist_m: double
child 3, invalid_initial_target_threshold_m: double
child 4, planned_count_after: int64
child 5, planned_count_before: int64
source_json: string
run_model: string
to
{'run_model': Value('string'), 'sanitization': {'dropped_invalid_initial_targets': Value('int64'), 'executed_count': Value('int64'), 'initial_target_to_exec0_dist_m': Value('float64'), 'invalid_initial_target_threshold_m': Value('float64'), 'planned_count_after': Value('int64'), 'planned_count_before': Value('int64')}, 'source_json': Value('string'), 'uid': Value('string'), 'validation': {'action_semantics': {'action_dim': Value('int64'), 'action_type': Value('string'), 'gripper_type': Value('string'), 'note': Value('string'), 'position_frame': Value('string'), 'raw_action_dim': Value('int64'), 'rotation_compose': Value('string'), 'rotation_frame': Value('string'), 'rotation_type': Value('string'), 'uses_controller_scaling': Value('bool')}, 'checkpoint_match_diagnostics': List({'checkpoint_index': Value('int64'), 'match_mode': Value('string'), 'position_error_m': Value('float64'), 'rotation_error_rad': Value('float64'), 'success': Value('bool')}), 'control_hz': Value('float64'), 'errors': List(Value('null')), 'executed_count': Value('int64'), 'executed_dense_count_for_ndtw': Value('int64'), 'executed_dense_count_full': Value('int64'), 'ok': Value('bool'), 'planned_count': Value('int64'), 'planned_dense_count_for_ndtw': Value('int64'), 'planned_dense_count_full': Value('int64'), 'policy_hz': Value('float64'), 'sim_hz': Value('float64'), 'tcp_site_name': Value('string'), 'warnings': List(Value('null'))}}
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
action_path: string
config_report: struct<dtw_penalty: string, exec_sr_pos_tol: double, exec_sr_rot_tol: double, executed_smoothness_pa (... 244 chars omitted)
child 0, dtw_penalty: string
child 1, exec_sr_pos_tol: double
child 2, exec_sr_rot_tol: double
child 3, executed_smoothness_path_length_normalized: bool
child 4, executed_smoothness_resample_points: int64
child 5, pos_tol: double
child 6, rot_tol: double
child 7, smoothing: struct<enabled: bool, method: null, params: null>
child 0, enabled: bool
child 1, method: null
child 2, params: null
child 8, tracking_ndtw_max_points: int64
child 9, tracking_ndtw_orientation_fallback: string
correction: struct<correction_exhausted_count: int64, correction_exhausted_rate: double, correction_steps: struc (... 103 chars omitted)
child 0, correction_exhausted_count: int64
child 1, correction_exhausted_rate: double
child 2, correction_steps: struct<count: int64, max: double, mean: double, median: double, p95: double>
child 0, count: int64
child 1, max: double
child 2, mean: double
child 3, median: double
child 4, p95: double
child 3, total_correction_steps: int64
cosmos_policy: struct<checkpoint_match_policy: string, control_hz: double, exec_sr_pos_tol: double, exec_sr_rot_tol (... 271 chars omitted)
child 0, checkpoint_match_policy: string
child 1, control_hz: double
child 2, exec_sr_pos_tol: double
child 3, exec_sr_rot_tol: double
child 4, ndtw_max_
...
uses_controller_scaling: bool
child 1, checkpoint_match_diagnostics: list<item: struct<checkpoint_index: int64, match_mode: string, position_error_m: double, rotation_er (... 32 chars omitted)
child 0, item: struct<checkpoint_index: int64, match_mode: string, position_error_m: double, rotation_error_rad: do (... 20 chars omitted)
child 0, checkpoint_index: int64
child 1, match_mode: string
child 2, position_error_m: double
child 3, rotation_error_rad: double
child 4, success: bool
child 2, control_hz: double
child 3, errors: list<item: null>
child 0, item: null
child 4, executed_count: int64
child 5, executed_dense_count_for_ndtw: int64
child 6, executed_dense_count_full: int64
child 7, ok: bool
child 8, planned_count: int64
child 9, planned_dense_count_for_ndtw: int64
child 10, planned_dense_count_full: int64
child 11, policy_hz: double
child 12, sim_hz: double
child 13, tcp_site_name: string
child 14, warnings: list<item: null>
child 0, item: null
sanitization: struct<dropped_invalid_initial_targets: int64, executed_count: int64, initial_target_to_exec0_dist_m (... 110 chars omitted)
child 0, dropped_invalid_initial_targets: int64
child 1, executed_count: int64
child 2, initial_target_to_exec0_dist_m: double
child 3, invalid_initial_target_threshold_m: double
child 4, planned_count_after: int64
child 5, planned_count_before: int64
source_json: string
run_model: string
to
{'run_model': Value('string'), 'sanitization': {'dropped_invalid_initial_targets': Value('int64'), 'executed_count': Value('int64'), 'initial_target_to_exec0_dist_m': Value('float64'), 'invalid_initial_target_threshold_m': Value('float64'), 'planned_count_after': Value('int64'), 'planned_count_before': Value('int64')}, 'source_json': Value('string'), 'uid': Value('string'), 'validation': {'action_semantics': {'action_dim': Value('int64'), 'action_type': Value('string'), 'gripper_type': Value('string'), 'note': Value('string'), 'position_frame': Value('string'), 'raw_action_dim': Value('int64'), 'rotation_compose': Value('string'), 'rotation_frame': Value('string'), 'rotation_type': Value('string'), 'uses_controller_scaling': Value('bool')}, 'checkpoint_match_diagnostics': List({'checkpoint_index': Value('int64'), 'match_mode': Value('string'), 'position_error_m': Value('float64'), 'rotation_error_rad': Value('float64'), 'success': Value('bool')}), 'control_hz': Value('float64'), 'errors': List(Value('null')), 'executed_count': Value('int64'), 'executed_dense_count_for_ndtw': Value('int64'), 'executed_dense_count_full': Value('int64'), 'ok': Value('bool'), 'planned_count': Value('int64'), 'planned_dense_count_for_ndtw': Value('int64'), 'planned_dense_count_full': Value('int64'), 'policy_hz': Value('float64'), 'sim_hz': Value('float64'), 'tcp_site_name': Value('string'), 'warnings': List(Value('null'))}}
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.
Dream.exe Benchmark
📄 Paper · 🤗 HF Daily Paper · 💻 Code · 🧠 DVD checkpoints
This release preserves the original 101 benchmark cases. The Dream.exe authors constructed the task instances, frozen scenes, camera settings, generation inputs, and reference data using RoboCasa data and environments.
Download and reproduce
After completing the code repository's installation guide, run from the code repository root:
hf download kaimingyang/Dream.exe --repo-type dataset \
--include 'bench/**' 'results/runs/**' 'results/videos/**' 'results/experiments/**' \
--local-dir data
hf download kaimingyang/DVD_for_Dream.exe \
--include 'DVD/lora/shared/*' 'DVD/lora/specific/*' \
--local-dir checkpoints
python -m dream_exe doctor --workspace configs/workspace.json
python -m dream_exe reproduce --workspace configs/workspace.json --dry-run
Check that the dry run lists the expected released runs: it lists descriptors only and does not validate their videos or execute the pipeline. An empty list means the result-input release is absent or the download is incomplete. Run one listed ID first, then the full matrix:
python -m dream_exe reproduce --workspace configs/workspace.json \
--run-id <ID_FROM_DRY_RUN> --stop-on-failure
python -m dream_exe reproduce --workspace configs/workspace.json
The download above is sufficient for recomputing results from frozen inputs
when all referenced files are available. Each generated-video directory must
include video.json, its original video, and its declared preprocessed video
(if any), with matching sizes and hashes. Run selections must refer to the
released benchmark collection or explicit released case UIDs.
Previously computed results/experiments/ bundles are included in the download
for inspecting or aggregating published results without rerunning. They are
not required by the recomputation path in reproduce.
Each such bundle requires its result.json, request.json, bound
resolved_config.json, and every artifact listed in result.json; filtering
down to metrics alone does not preserve result integrity. The matching
results/runs/ descriptors are also needed for run discovery and aggregation.
The complete benchmark guide, including provider setup and execution options,
is BENCHMARK.md.
Use --revision <commit> on HF downloads to pin an experiment. Default paths
are data/bench, data/results, and checkpoints/DVD/lora; change workspace
roots and bindings if placing assets elsewhere.
Data layout
bench/: the 101 cases, frozen initialization, prompts, protocols, and GT video/action/depth references.results/runs/: configurations for reproducing the released routes.results/videos/: generated videos and policy-rollout videos consumed by trajectory extraction.results/experiments/: previously released trajectory, execution, and evaluation artifacts, when available.
New runs write to the code workspace's output directory. GT video with model depth and GT video with reference depth are separate reference routes. CosmosPolicy inputs are videos rendered from policy actions; Dream.exe extracts trajectories from those videos and executes the extracted actions.
Unavailable generation inputs
Some requested videos could not be generated because of the generation
models' compliance restrictions, as reported by the benchmark authors. These
entries are unavailable inputs and are recorded as input_missing; affected
runs return a partial report rather than a fully completed report. Availability
is separate from execution success. A later reconstructed video is identified
separately and does not establish the identity of the original input.
Acknowledgements
We thank RoboCasa for the source data and simulation environments, and DVD for the depth foundation model used by our benchmark-fine-tuned checkpoints. Third-party assets and model outputs retain the terms recorded by their sources. Please cite Dream.exe using the code repository's CITATION.cff.
Optional pickled depth-cache metadata is excluded and listed in results/publication-omissions.json; numerical depth arrays are preserved. This publication supports reading results and recomputation, not reuse of the original private depth cache.
Included results
The release includes 101 benchmark cases and 2,404 stored experiment bundles across 24 routes. Veo3.1 enhanced includes 101 videos and 101 result bundles. Existing results are retained, with 18 supplementary Veo3.1 enhanced results. Numerical artifacts and media are preserved; publication identifiers and portable locations have corresponding integrity seals.
Rights fields retain their imported values, including unknown or null entries; they have not been converted into affirmative licenses. Publication was authorized by the dataset owner. Applicable original third-party terms and provenance remain attached. Producer-side evidence locators in supplementary evaluation plans identify original inputs; an unattached locator is not a claim that the source file is included or that the plan can be replayed directly.
- Downloads last month
- 155
Models trained or fine-tuned on kaimingyang/Dream.exe