Datasets:

yangkaixing
/

FlashMotion-all-predictions-20261016

Error code:   DatasetGenerationError
Exception:    UnicodeDecodeError
Message:      'utf-8' codec can't decode byte 0xd2 in position 16: invalid continuation byte
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1827, in _prepare_split_single
                  for key, table in generator:
                                    ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
                  for item in generator(*args, **kwargs):
                              ~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/text/text.py", line 98, in _generate_tables
                  batch = f.read(self.config.chunksize)
                File "/usr/local/lib/python3.14/site-packages/datasets/utils/file_utils.py", line 839, in read_with_retries
                  out = read(*args, **kwargs)
                File "<frozen codecs>", line 325, in decode
              UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd2 in position 16: invalid continuation byte
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1694, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1879, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

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.

text string
Ablation folder labels / 论文消融名称对照

GenerativeStrategy_1step / GenerativeStrategy_4step
Ours_vs_Diffusion = Ours vs Diffusion
Ours_vs_FlowMatching = Ours vs Flow Matching

TrainingObjective_I
MeanFlow_only = MeanFlow only（真实NaN保留，并非可正常绑定的人体动作）
with_Uncond = w/ Uncond
FlowMatching = w/o MeanFlow loss
注意：这个Objective里的FlowMatching使用无MeanFlow损失的原始x0-FlowMatching
控制模型；不是额外漏掉了一个without_MeanFlow_loss文件夹。

TrainingObjective_II
without_Kinematic = w/o Kinematic loss（仅对应FK项去除，其他保留项按原实验）
without_Dynamic = w/o Dynamic loss（原实验的velocity/acceleration消融）

MotionBackbone
Mamba, Transformer_NonAR, Transformer_Streamline = 三个原始骨干消融。
使用各自已选checkpoint及原生wrapper，没有为展示修改其动作。

ConditionFusion
AdaIN_only = AdaIN only
Scan_only = Scan only

TimeSampling_4step
Ours = 已发布Uniform 4-step，epoch5850。
Beta_0.5_0.5 = Beta(0.5,0.5)，用户最后指定epoch5150。
Beta_2_2 = Beta(2,2)，epoch5500。
Beta_1_2 = Beta(1,2)，epoch1500。
Beta_2_1 = Beta(2,1)，epoch2100。
每个Beta都用其对应分布的固定分位数时间网格，而非拿Uniform网格冒充。

同一章节内可能有多个样本池：原始候选、独立100候选、额外simple候选、完整311。
这些是不同文本/seed的真实池，不是重复复制一个最终视频片段。
每个1v1目录里的Ours和对手按同名文件匹配。详细来源和checkpoint见同名JSON。
NPZ collections by folder. Counts are exact after the COMPLETE coverage gate.
Each independent_single_bundle file can contain multiple separately sampled segments.

Ablation/ConditionFusion/Ours_vs_AdaIN_only/BABEL_val311/AdaIN_only 311
Ablation/ConditionFusion/Ours_vs_AdaIN_only/BABEL_val311/Ours 311
Ablation/ConditionFusion/Ours_vs_AdaIN_only/ablation_five_pools_fusion/AdaIN_only 100
Ablation/ConditionFusion/Ours_vs_AdaIN_only/ablation_five_pools_fusion/Ours 100
Ablation/ConditionFusion/Ours_vs_AdaIN_only/video_legacy100/AdaIN_only 100
Ablation/ConditionFusion/Ours_vs_AdaIN_only/video_legacy100/Ours 100
Ablation/ConditionFusion/Ours_vs_Scan_only/BABEL_val311/Ours 311
Ablation/ConditionFusion/Ours_vs_Scan_only/BABEL_val311/Scan_only 311
Ablation/ConditionFusion/Ours_vs_Scan_only/ablation_five_pools_fusion/Ours 100
Ablation/ConditionFusion/Ours_vs_Scan_only/ablation_five_pools_fusion/Scan_only 100
Ablation/ConditionFusion/Ours_vs_Scan_only/video_legacy100/Ours 100
Ablation/ConditionFusion/Ours_vs_Scan_only/video_legacy100/Scan_only 100
Ablation/GenerativeStrategy_1step/Ours_vs_Diffusion/ablation_five_pools_strategy/Diffusion 100
Ablation/GenerativeStrategy_1step/Ours_vs_Diffusion/ablation_five_pools_strategy/Ours 100
Ablation/GenerativeStrategy_1step/Ours_vs_Diffusion/video_legacy100/Diffusion 100
Ablation/GenerativeStrategy_1step/Ours_vs_Diffusion/video_legacy100/Ours 100
Ablation/GenerativeStrategy_1step/Ours_vs_FlowMatching/ablation_five_pools_strategy/FlowMatching 100
Ablation/GenerativeStrategy_1step/Ours_vs_FlowMatching/ablation_five_pools_strategy/Ours 100
Ablation/GenerativeStrategy_1step/Ours_vs_FlowMatching/video_legacy100/FlowMatching 100
Ablation/GenerativeStrategy_1step/Ours_vs_FlowMatching/video_legacy100/Ours 100
Ablation/GenerativeStrategy_4step/Ours_vs_Diffusion/BABEL_val311/Diffusion 311
Ablation/GenerativeStrategy_4step/Ours_vs_Diffusion/BABEL_val311/Ours 311
Ablation/GenerativeStrategy_4step/Ours_vs_Diffusion/video_legacy100/Diffusion 100
Ablation/GenerativeStrategy_4step/Ours_vs_Diffusion/video_legacy100/Ours 100
Ablation/GenerativeStrategy_4step/Ours_vs_Diffusion/video_strategy100_4step/Diffusion 100
Ablation/GenerativeStrategy_4step/Ours_vs_Diffusion/video_strategy100_4step/Ours 100
Ablation/GenerativeStrategy_4step/Ours_vs_FlowMatching/BABEL_val311/FlowMatching 311
Ablation/GenerativeStrategy_4step/Ours_vs_FlowMatching/BABEL_val311/Ours 311
Ablation/GenerativeStrategy_4step/Ours_vs_FlowMatching/video_legacy100/FlowMatching 100
Ablation/GenerativeStrategy_4step/Ours_vs_FlowMatching/video_legacy100/Ours 100
Ablation/GenerativeStrategy_4step/Ours_vs_FlowMatching/video_strategy100_4step/FlowMatching 100
Ablation/GenerativeStrategy_4step/Ours_vs_FlowMatching/video_strategy100_4step/Ours 100
Ablation/MotionBackbone/Ours_vs_Mamba/BABEL_val311/Mamba 311
Ablation/MotionBackbone/Ours_vs_Mamba/BABEL_val311/Ours 311
Ablation/MotionBackbone/Ours_vs_Mamba/ablation_five_pools_backbone/Mamba 100
Ablation/MotionBackbone/Ours_vs_Mamba/ablation_five_pools_backbone/Ours 100
Ablation/MotionBackbone/Ours_vs_Transformer_NonAR/BABEL_val311/Ours 311
Ablation/MotionBackbone/Ours_vs_Transformer_NonAR/BABEL_val311/Transformer_NonAR 311
Ablation/MotionBackbone/Ours_vs_Transformer_NonAR/ablation_five_pools_backbone/Ours 100
Ablation/MotionBackbone/Ours_vs_Transformer_NonAR/ablation_five_pools_backbone/Transformer_NonAR 100
Ablation/MotionBackbone/Ours_vs_Transformer_Streamline/BABEL_val311/Ours 311
Ablation/MotionBackbone/Ours_vs_Transformer_Streamline/BABEL_val311/Transformer_Streamline 311
Ablation/MotionBackbone/Ours_vs_Transformer_Streamline/ablation_five_pools_backbone/Ours 100
Ablation/MotionBackbone/Ours_vs_Transformer_Streamline/ablation_five_pools_backbone/Transformer_Streamline 100
Ablation/TimeSampling_4step/Ours_vs_Beta_0.5_0.5/BABEL_val311/Beta_0.5_0.5 311
Ablation/TimeSampling_4step/Ours_vs_Beta_0.5_0.5/BABEL_val311/Ours 311
Ablation/TimeSampling_4step/Ours_vs_Beta_1_2/BABEL_val311/Beta_1_2 311
Ablation/TimeSampling_4step/Ours_vs_Beta_1_2/BABEL_val311/Ours 311
Ablation/TimeSampling_4step/Ours_vs_Beta_2_1/BABEL_val311/Beta_2_1 311
Ablation/TimeSampling_4step/Ours_vs_Beta_2_1/BABEL_val311/Ours 311
Ablation/TimeSampling_4step/Ours_vs_Beta_2_2/BABEL_val311/Beta_2_2 311
Ablation/TimeSampling_4step/Ours_vs_Beta_2_2/BABEL_val311/Ours 311
Ablation/TrainingObjective_I/Ours_vs_FlowMatching/BABEL_val311/FlowMatching 311
Ablation/TrainingObjective_I/Ours_vs_FlowMatching/BABEL_val311/Ours 311
Ablation/TrainingObjective_I/Ours_vs_FlowMatching/ablation_five_pools_objective_i/FlowMatching 100
Ablation/TrainingObjective_I/Ours_vs_FlowMatching/ablation_five_pools_objective_i/Ours 100
Ablation/TrainingObjective_I/Ours_vs_FlowMatching/objective_i_simple_pool_r1_objective_i/FlowMatching 100
Ablation/TrainingObjective_I/Ours_vs_FlowMatching/objective_i_simple_pool_r1_objective_i/Ours 100
Ablation/TrainingObjective_I/Ours_vs_MeanFlow_only/BABEL_val311/MeanFlow_only 311
Ablation/TrainingObjective_I/Ours_vs_MeanFlow_only/BABEL_val311/Ours 311
Ablation/TrainingObjective_I/Ours_vs_MeanFlow_only/ablation_five_pools_objective_i/MeanFlow_only 100

End of preview.

⚡ FlashMotion — Motion Predictions

Native, denormalized predictions for character animation and retargeting. This is a numerical handoff, not a video, model-weight or benchmark-leaderboard release.

📦 Download

Download FlashMotion-all-predictions-20261016-final.zip (11.25 GB), then unzip it. The complete archive contains 26,130 prediction files, organized from 19,243 canonical records, with 7,576 matched 1v1 ablation pairs. Repeated Ours files in pair folders are intentional.

Task	Native output	Frame rate
Text-to-Motion	Physical HumanML3D-263 + 22-joint positions	20 fps
Speech-to-Gesture	Native SMPL-X 55-joint axis-angle poses + translation, shape and expression	30 fps
Music-to-Dance	Native 24/22-joint positions; SMPL body poses where actually saved	30 fps

🗂️ Layout

FlashMotion-all-predictions-20261016/
├── Comparison/     # T2M, S2G, M2D: full cohorts + previous candidate pools
├── Ablation/       # Matched Ours_vs_Other folders, all samples
├── Extras/         # Independent T2M single-segment predictions
├── Inputs/         # Matched music/speech and available transcripts
├── 使用说明.txt
├── *_FORMAT.txt    # Dimensions, units, axes and conventions
└── load_prediction.py

🚀 Use

python -m pip install numpy

Load NPZ files with numpy.load(path, allow_pickle=False). Read 使用说明.txt and the matching *_FORMAT.txt before retargeting. T2M-263 is not SMPL pose parameters; M2D Ours full-cohort output is joints-only where no original pose parameters were saved. Original native method post-processing is retained; no additional presentation IK, smoothing or fabricated poses were added for this handoff.

Original failed MeanFlow-only NaNs and genuine extreme outputs are preserved and listed in ACTUAL_FAILURES_AND_OUTLIERS.json; they are not missing files or repaired animations. BABEL full-cohort files are the 311-episode validation split; custom caption pools are separately named.

SHA256: e5d59993e1365c95996d58ee38a3516396f52c0b8dcebc420f2d34e1e7472fe4. Per-file hashes, exact coverage, checkpoint/source bindings and the Chinese handoff guide are inside the ZIP. See DELIVERY.json for the archive checksum and counts. For this collaboration handoff only; underlying source-data terms remain applicable.

Downloads last month: 81

Total file size:

11.3 GB