Dataset Viewer
The dataset viewer is not available for this subset.
Cannot get the split names for the config 'default' of the dataset.
Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/hdf5/hdf5.py", line 49, in _split_generators
                  import h5py
              ModuleNotFoundError: No module named 'h5py'
              
              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/split_names.py", line 71, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

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.

TacSpike Slip Detection 1kHz

数据集简介

这是 TacSpike 项目的 1kHz 视触觉事件数据集发布版,任务为 滑移二分类。数据由原始触觉帧/视频经过 v2e + SuperSloMo 生成 1kHz DVS-like 事件,再按 20ms window 和 1ms stride 整理成 sequence HDF5。

dataset: tacspike-slip-detection-1khz
version: v1.0.0
task: slip_binary
event_source: v2e_superslomo_1khz
timestamp_resolution_ms: 1.0
window_ms: 20.0
stride_ms: 1.0
bins: 20
voxel_shape: [20, 2, 128, 128]
source_dataset: TacBench/Sparsh tactile frames

文件结构

README.md                         # Hugging Face dataset card
README_DATA.md                    # 数据包内说明
VERSION
metadata.json
SHA256SUMS.txt
summary.json
manifest_sequences.csv
manifest_windows.csv
verification_report.json
baseline_report.json
DATA_LICENSE
LICENSE
CITATION.cff
examples/
  inspect_h5_schema.py
  quickstart_slip.py
  quickstart_marker.py
sequences/
  train/{sequence_id}.h5
  val/{sequence_id}.h5
  test/{sequence_id}.h5

数据规模

num_sequences: 1559
num_sequences_by_split: {'train': 1091, 'val': 234, 'test': 234}
num_windows: 28952409
split_counts: train: 20293980 / val: 4173170 / test: 4485259
mean_event_count_per_window: 2.989931304161944
empty_window_ratio: 0.3324946811852513

类别统计

num_slip_windows: 7942685
num_no_slip_windows: 21009724
class_ratio_slip: 0.2743358937765766

HDF5 结构

每个 sequence 文件是一个独立 HDF5 文件,根属性中包含 sequence_idformatheightwidthbinswindow_msstride_ms 等字段。

events/
  t                         # float64, 秒,1ms 网格上的事件时间戳
  x                         # int, [0, width)
  y                         # int, [0, height)
  p                         # int, 0/1 polarity
windows/
  t_start                   # 每个训练窗口起点
  t_end                     # 每个训练窗口终点
  t_label                   # 该窗口对应的标签时刻
  event_count               # 该窗口内事件数
label/
    slip

默认输入语义是:对每个 windows[i],取 [t_start[i], t_end[i]] 内的事件,动态 voxelize 成 (20, 2, 128, 128);窗口长度为 20ms,stride 为 1ms。

快速使用

安装最小依赖:

python -m pip install numpy h5py

在下载后的数据集根目录运行:

python examples/inspect_h5_schema.py --data-root .
python examples/quickstart_slip.py --data-root .

如果在代码仓库中使用完整工具链,可以运行:

python scripts/verify_dataset.py --input /path/to/dataset --task slip_binary
python scripts/train_baseline.py --input /path/to/dataset --task slip_binary --max-train-samples 2000 --max-eval-samples 500 --epochs 1

Baseline

baseline_report.json 中记录的是轻量 CNN baseline,主要用于验证数据读取、label 对齐和训练流程,不代表最终 SNN 性能。

{
  "cnn": {
    "accuracy": 0.2776,
    "f1": 0.4345648090169067,
    "precision": 0.2776,
    "recall": 1.0
  },
  "device": "cuda",
  "epochs": 1,
  "eval_samples": 5000,
  "majority_baseline": {
    "accuracy": 0.7224,
    "f1": 0.0,
    "precision": 0.0,
    "recall": 0.0
  },
  "sequence_input": true,
  "task": "slip_binary",
  "train_samples": 20000
}

校验

下载或迁移数据后建议先做校验:

sha256sum -c SHA256SUMS.txt
python examples/inspect_h5_schema.py --data-root .

已知限制

  • 滑移类别不均衡,发布版应同时报告 precision、recall、F1 和 PR-AUC。

  • 当前 CNN baseline 用于验证数据读取和训练流程,不代表最终 SNN 性能。

  • 本数据集是由现有触觉数据集派生得到的事件数据,正式公开使用时应同时遵守原始数据集、v2eSuperSloMo 及本发布包的许可要求。

  • 当前数据标签和事件均按 1ms 时间栅格组织,适合 1kHz 级滑移检测或触觉运动估计实验;如果模型使用不同采样频率,需要显式重新定义 window/stride。

许可

本数据集使用 license: other。数据文件是派生数据,二次分发和使用须遵守原始数据集及相关工具/模型权重的许可。详见 DATA_LICENSELICENSE。在原始数据许可完全确认之前,不应把本发布包标成 MIT、Apache、CC-BY 或 CC0。

引用

请引用 CITATION.cff 中的 TacSpike 数据集条目,并同时引用对应的原始数据集、v2eSuperSloMo。后续论文/DOI 确定后,应更新本 dataset card、CITATION.cff 和 GitHub README。

Downloads last month
54