v0.1.0 Research Preview

v0.1.0 Research Preview

This prerelease freezes the public research pipeline at Git commit
5b53f7931a39a89bde4e7906b9ababf3a6c53ee4 and provides one citable checkpoint for the website, browser
demo, Android preview, model manifest, and reproduction workflow.

Frozen artifacts

• Package version: imu-activity-pipeline 0.1.0 on Python 3.12 or newer.
• Public model repository: config-h/Wearable-IMU-Activity-Segmentation-Pipeline
  at revision main.
• Model manifest: model-assets.json, schema version 1, with per-file sizes and
  SHA-256 hashes for the PyTorch and ONNX assets.
• Manifest source revision: 3f757218da0e8ce8870f0e12f7994c28bbe31f9a.
• Browser demo: Gradio 6.25.0 with an explicit English / Simplified Chinese
  switch and a bundled, deterministic 120-second synthetic recording.

Browser demo operating point

• Fusion: local_boundary
• Minimum activity duration: 5 seconds
• Confidence threshold: 0.30
• Top-K records: 5
• Input contract: seven tab-separated columns at approximately 100 Hz, with
  800–60,000 valid samples and a 20 MB upload limit

Independent external-test operating point

The frozen LBSA + TRL pipeline reports 0.89 mean-user F1 and 0.90 micro-F1 on
37 independent recordings containing 114 labeled segments. Same-class records
are matched one-to-one at IoU > 0.5. External labels were not used to select the
checkpoints, fusion rule, or Temporal Record Layer policy.

Environment

The reference environment uses Python 3.12+, NumPy 1.26.4, pandas 2.2.3,
SciPy 1.15.3, Matplotlib 3.9.2, scikit-learn 1.5.2, PyTorch 2.8.0,
Gradio 6.25.0, and OpenPyXL 3.1.5. Exact declarations remain in
environment.yml, requirements.txt, and pyproject.toml.

Known limitations

• Participant recordings are access-controlled and are not included in this
  repository or release.
• The six-class model was developed for the documented wrist-worn sensing
  protocol; new devices, placements, populations, or activities require new
  validation.
• The synthetic Space sample demonstrates the full input-to-output path but is
  not accuracy evidence.
• Predictions are research outputs and are not medical, safety, or coaching
  advice.
• The model repository revision is currently main; integrity is frozen by the
  manifest's per-asset SHA-256 hashes rather than by a Hub commit identifier.

Android demo APK v1.0 preview (arm64-v8a)

Android 体验版 APK（arm64-v8a）

适用手机类型：

• 适用于 64 位 ARM Android 手机，设备 ABI 需要支持 arm64-v8a。
• 大多数近年的 Android 手机属于这一类，包括使用 Qualcomm Snapdragon、MediaTek Dimensity/Helio、Samsung Exynos、Google Tensor、HiSilicon Kirin 等 64 位 ARM 芯片的机型。
• 系统要求：Android 7.0 或更高版本（minSdk/API 24；targetSdk/API 34）。
• 不适用于 iPhone/iOS。
• 不适用于 x86/x86_64 Android 模拟器。
• 不适用于仅支持 32 位 armeabi-v7a 的较老 Android 设备。

APK 信息：

• 文件：hls-har-android-demo-v1.0-arm64-v8a-debug.apk
• 包名：com.imu.realtime
• 应用名：IMU 实时数据 / IMU Realtime
• 版本：1.0 (versionCode=1)
• Native ABI：arm64-v8a
• 大小：约 33 MB
• 签名：Android Debug certificate；这是体验/测试版，不是正式生产签名版本。
• SHA256：cdde56db9d915eb10918724d503597a84fb18deace096086fe87509f60348be6

使用提示：

• Android 12 及以上通常需要授予附近设备/蓝牙权限；部分系统还会要求定位权限才能扫描 BLE 设备。
• 该体验版面向 WT9011DCL-BT50 BLE IMU 采集与本地 ONNX 推理流程验证。