The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: RuntimeError
Message: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.
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 2383, in __iter__
for key, example in self.ex_iterable:
^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 315, in __iter__
for key_example in islice(self.generate_examples_fn(**gen_kwargs), shard_example_idx_start, None):
~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 127, in _generate_examples
for example_idx, example in enumerate(self._get_pipeline_from_tar(tar_path, tar_iterator)):
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 54, in _get_pipeline_from_tar
current_example[field_name] = cls.DECODERS[data_extension](current_example[field_name])
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 332, in torch_loads
return torch.load(io.BytesIO(data), weights_only=True)
~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 1601, in load
return _load(
opened_zipfile,
...<3 lines>...
**pickle_load_args,
)
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 2221, in _load
result = unpickler.load()
File "/usr/local/lib/python3.14/site-packages/torch/_weights_only_unpickler.py", line 541, in load
self.append(self.persistent_load(pid))
~~~~~~~~~~~~~~~~~~~~^^^^^
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 2185, in persistent_load
typed_storage = load_tensor(
dtype, nbytes, key, _maybe_decode_ascii(location)
)
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 2147, in load_tensor
wrap_storage = restore_location(storage, location)
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 734, in default_restore_location
result = fn(storage, location)
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 667, in _deserialize
device = _validate_device(location, backend_name)
File "/usr/local/lib/python3.14/site-packages/torch/serialization.py", line 634, in _validate_device
raise RuntimeError(
...<5 lines>...
)
RuntimeError: Attempting to deserialize object on a CUDA device but torch.cuda.is_available() is False. If you are running on a CPU-only machine, please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.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.
PhysDelta
Numerical physical control for video generation
PhysDelta evaluates how target and non-target objects respond to physical interventions. This repository contains evaluation inputs and a separate raw-training-data distribution for EnactPhys.
Project website · Code and evaluation · Checkpoints
Benchmark inputs
The input release contains all tasks in the following benchmark tracks. Each task specifies its input image, physical-condition file, initial tracking masks, prompt, seed and frame rate.
| Track | Tasks | Cases | Seeds |
|---|---|---|---|
| Sim parameter control | 2,385 | 71 | 42, 3407, 918273 |
| Sim invariance | 60 | 10 | 42, 3407, 918273 |
| Sim object control | 201 | 6 | 42, 3407, 918273 |
| Real parameter control | 804 | 103 | 3407, 424242, 918273 |
| Real object control | 69 | 5 | 3407, 424242, 918273 |
The control tracks and additional Real quality tasks contain 3,594 distinct tasks. The full release includes 189 input images, 189 initial tracking-mask files and 1,193 physical-condition files. Shared inputs are stored once. Six additional Real friction-diagnostic tasks are provided separately and are not included in the formal object-control aggregate. The Real parameter-control inventory includes 708 matched comparisons.
real/quality.csv specifies the fixed 100-video quality/plausibility panel: 75 additional Parameter Control tasks with seed 42 and 25 tasks shared with Object Control. These 75 additional tasks bring Real to 948 distinct tasks and the combined benchmark to 3,594. sim/plausibility.csv identifies the 215-video simulation plausibility panel; sim/video_quality.csv identifies the 215-video Video Quality panel. Their union includes 18 quality-only task IDs outside the current control-track manifests. These inputs are retained explicitly to reproduce the scored panels, bringing the released evaluation-task union to 3,612, plus the six separate friction diagnostics. Input images and conditions are shared with the corresponding numeric-control tasks.
Download benchmark inputs
hf download EnactPhys/PhysDelta --repo-type dataset --exclude 'training/*' 'training_raw/*' 'training_sources/*' --local-dir data/PhysDelta
cd data/PhysDelta
tar -xzf sim_inputs.tar.gz
tar -xzf real_inputs.tar.gz
Each archive contains a sim/ or real/ directory with task CSVs, images/, conditions/ and masks/. All paths in the CSVs are relative to the extracted dataset root. video denotes the input image, preserving the generation loader interface. The original input image is used; reference or generated future video frames are not required as model inputs.
Condition files use PyTorch serialization and preserve the original numerical tensors. They include initial object masks and physical parameters; the inference loader selects its defined conditioning fields. Initial tracking masks are stored as NumPy .npz files with a segmentation array. They are used for video measurement, separately from latent-resolution model conditioning.
benchmark.json provides the inventory counts. The CSV copies under manifests/ can be inspected without extracting the archives. Seeds and task IDs are fixed for repeatable task selection; numerical reproducibility also depends on the matching model, sampling configuration and evaluation environment.
Physics-IQ inputs
physicsiq_inputs.tar.gz contains 114 input images and 80 physical-condition files for the fixed Solid Mechanics configuration. The code repository specifies all 342 tasks: 114 views × three seeds, with adapter/base routing fixed per view.
hf download EnactPhys/PhysDelta physicsiq_inputs.tar.gz --repo-type dataset --local-dir data/PhysDelta
tar -xzf data/PhysDelta/physicsiq_inputs.tar.gz -C data/PhysDelta
Official reference videos and masks are obtained from Physics-IQ Verified. Follow the generation-and-scoring command.
Evaluation
Use the code repository for model setup, generation and evaluation. Video workflows consume generated evaluation videos; recorded aggregation consumes the published measurements and judgments. Use the fixed quantitative manifests rather than the website's selected examples.
Training data
Raw training data contains original MP4 videos, physical-condition tensors and fixed train/validation manifests (109,800 / 4,196 rows). The coverage and shard index records upload progress and missing original-video references. The training-data guide provides download, extraction, preprocessing and cache-binding commands.
hf download EnactPhys/PhysDelta --repo-type dataset --include 'training_raw/*' --local-dir data/PhysDelta
Additional original source videos are mapped to 15,310 rows by sample identifier. Their historical trimming/reencoding equivalence remains to be verified. These are separate from the exact located training-video files.
Previously uploaded optional encoding caches are retained under training/; this subset is not required for the raw-data download and is not a complete cache distribution. Third-party benchmarks retain their own distributions and protocols.
MORPHEUS video tracking and scoring and screening are documented in the code repository.
Licenses
Third-party source datasets retain their licenses and attribution. A project-specific license for the EnactPhys/PhysDelta contribution has not yet been assigned. Official third-party benchmark references remain available from their original distributors.
- Downloads last month
- 243