v0.13.0 - Motor Primitive Annotation System

🦖 v0.13.0 — Motor Primitive Annotation System

基于 T-Rex 论文定义的 22 个标准 Motor Primitive，为触觉数据提供 primitive 级别标注。

✨ New Features

• PrimitiveAnnotation 类 — 22 个标准 Motor Primitive（wrap, lift, grasp, fold, cut, insert, press, wipe, peel, assemble, extract, twist, shake, dispense, disassemble, squeeze, pour, open, close, screw, unscrew, reach）
• Primitive 时间轴轨道 — Panel UI 中新增彩色 primitive 轨道，在 phase 轨道下方显示
• AI 预标注 — 基于力/接触信号的规则推断 primitive（PredictEngine.predict_primitives()）
• 导出支持 — JSON 新增 primitive_annotations 字段，CSV 新增 primitive_label 列
• Demo 数据 — tlabel.demo('primitives_demo') 展示 reach→grasp→lift→wrap→press 序列
• 向后兼容 — 旧版本 JSON 正常加载，无 primitive_annotations 字段时不影响

📦 API 变更

# 新增 API
import tlabel

# 添加 primitive 标注
data.add_primitive('grasp', start_frame=10, end_frame=30, confidence=0.95)

# 获取 primitive 时间线
timeline = data.get_primitive_timeline()

# AI 预标注 primitive
engine = tlabel.PredictEngine()
engine.apply_primitives(data, min_confidence=0.4)

# Demo
data = tlabel.demo('primitives_demo')
data.review()  # 显示 primitive 轨道

📁 Changed Files

• tlabel/core/primitive.py (NEW) — PrimitiveAnnotation 类 + 22 primitive 预设
• tlabel/core/types.py — 新增 primitive_label, primitive_confidence, primitive_annotations
• tlabel/viewer/templates.py — Primitive 时间轴轨道 + 颜色映射
• tlabel/predict/engine.py — predict_primitives(), apply_primitives()
• tlabel/export/writer.py — CSV primitive_label 列
• tlabel/adapters/tlabel_format.py — 向后兼容加载 primitive_annotations
• tlabel/demo.py — primitives_demo
• tlabel/_version.py, pyproject.toml — v0.13.0

v0.12.4 - 修复gelsight_images demo数据格式

v0.12.4 修复

🐛 Bug修复

• demo('gelsight_images') 返回0帧：demo_gelsight_images.json 的帧数据嵌套在 episodes[0].frames 中，但demo函数期望顶层 frames 字段
• 修复JSON格式：帧数据移至顶层，使用正确的 tlabel_v2 结构
• 合成图像正常生成（10帧 × 320×240灰度图）

✅ 验证

• tlabel.demo('gelsight_images') → 10帧 + 10张合成触觉图像
• 面板正确显示图像序列
• 版本号显示 v0.12.4

📦 安装

pip install tlabel==0.12.4 --upgrade

v0.12.3 - 面板版本号动态化

v0.12.3 修复

🐛 Bug修复

• 面板版本号硬编码问题：templates.py 中面板标题的版本号写死为 v0.8.0，导致无论安装什么版本，面板始终显示 v0.8.0
• 修复为动态读取 tlabel._version.__version__，面板版本号随包版本自动更新

📦 发布

• PyPI: https://pypi.org/project/tlabel/0.12.3/
• 安装: pip install tlabel==0.12.3 --upgrade

v0.12.2 - 图像可视化 & Demo修复

v0.12.2

• 修复 tlabel.demo() 未解析 image/image_path 字段的问题
• 新增 tlabel.demo("gelsight_images") 自动生成合成触觉图像
• 面板右侧显示触觉图像序列（播放/暂停/进度条）
• 修复 tlabel/_version.py 版本同步问题

v0.12.0

• 新增图像可视化：面板右侧显示触觉图像序列
• TLabelFrame 支持 image / image_path 字段
• 新增 examples/demo_with_images/ 数据集

v0.11.2

• 修复 JS 语法转义 + Jupyter 初始化时序问题

v0.8.0 - FTP-1/MTTS Zarr Export

🚀 TLabel v0.8.0 — FTP-1 / MTTS Integration

New Features

• FTP-1 Converter (tlabel/converters/ftp1.py): Export labeled data to FTP-1 MTTS Zarr format
  • tlabel_to_ftp1() / batch_to_ftp1() core functions
  • 21 MTTS functional areas (15 hand zones + 6 wrist torque channels)
  • 7 registered sensors: GelSight, GelSightMini, FreeTacMan, ViTaMIn, 3DViTac, Contactile, BinaryContact
  • 4 preset mappings: parallel_gripper, three_finger, five_finger, dexterous_hand
  • image/matrix/binary modality support
  • Auto image resize 224×224 + normalization to float32 [-1, 1]
• data.export_ftp1(): One-line export from TLabelData
• 🚀 Export Tab in Panel: Sensor selection, functional area picker with presets, export preview + Python command generation
• Zarr backend: Append mode for multi-episode datasets

Installation

pip install tlabel[ftp1]   # installs zarr>=2.16

Quick Start

from tlabel import demo
data = demo("gelsight")
data.export_ftp1("output.zarr",
    sensor_name="GelSightMini",
    functional_areas=[0, 1])  # thumb tip + index fingertip

Compatibility

Exported Zarr files are directly compatible with FTP-1 for fine-tuning the world first general-purpose tactile foundation model.

v0.7.0 — Sensor Profile & Feature Metadata

What's New in v0.7.0

🆕 Sensor Profile (sensor_profile)

TLabelData now carries optional sensor metadata — elastomer properties, optical specs, and calibration status. This is the foundation for physics-aware calibration and the path toward TLabel as a sensor-agnostic metadata standard (like EXIF for images).

data.sensor_profile = {
    "sensor_type": "GelSight Mini",
    "elastomer": {"material": "PDMS", "modulus_pa": 1.8e6, "thickness_mm": 2.0},
    "optical": {"resolution": [320, 240], "fps": 30},
    "calibration": {"force_model": None, "status": "uncalibrated"}
}

🆕 Feature Metadata Registry (features_meta.py)

All 22+1 features now have static metadata: category, computation formula, physical semantics, SI units, force correlation level, and calibration dependencies. Transparent, self-documenting, no black boxes.

Categories:

• deformation (5): contact, deformation_magnitude, force_peak, force_direction, deformation_magnitude_peak
• gradient (4): slip_entropy, slip_event, texture_energy, edge_density
• force_semantic (9): contact_area, centroid_x, normal_field_magnitude, shear_field_, delta_force_, friction_cone_ratio
• temporal (4): optical_flow_*, temporal_deformation_rate, contact_transition

🆕 deformation_magnitude_peak

Honest replacement for the deprecated force_magnitude. Same computation, transparent naming — no more implying a force measurement in Newtons when the value is uncalibrated pixel intensity.

⚠️ Deprecation: force_magnitude

force_magnitude now emits a DeprecationWarning. It remains functional (alias of deformation_magnitude) but will be removed in v1.0. Migrate to deformation_magnitude_peak.

📦 Schema & Export

• to_dict() now outputs feature_metadata and sensor_profile sections
• schema_version bumped from 0.4.0 → 0.7.0
• Full backward compatibility: old JSON files load without errors

✅ Tests

• 121 tests, 119 passed
• 2 stale version assertions in legacy test files (test_v040, test_v050) — not functional regressions

---

Full Changelog: v0.6.2...v0.7.0

v0.6.2 - ToucHD Demo + Regression Test

What's New

ToucHD Demo Data

tlabel.demo('touchd') now works! 100-frame simulated press sequence.

import tlabel
data = tlabel.demo('touchd')
data.review()

Release Regression Test

tests/release_regression.py — 11 tests covering all adapters and demos.

python tests/release_regression.py

Bug Fixes

• Fix missing _version.py in wheel
• Fix missing submodules in wheel

Full Changelog: v0.5.3...v0.6.2

v0.6.1 - ToucHD-Force Adapter

What's New

ToucHD-Force Adapter (AnyTouch 2, ICLR 2026)

Full support for ToucHD-Force dataset from AnyTouch 2 (ICLR 2026) by GeWu-Lab + BAAI.

Features:

• 4 sensors: DIGIT, BioTip, GelSight, DuraGel
• 3D contact force ground truth (Fx, Fy, Fz) with sensor-specific normalization
• Action labels (press, slide, etc.) + left/right hand image selection
• 22-dim TLabel v2 feature extraction + force-driven slip detection
• Optical flow computation (Farneback, when cv2 available)
• Manipulation phase inference

Usage:

import tlabel

# Auto-detect
data = tlabel.load("ToucHD-Force/", sensor="gelsight")

# With params
data = tlabel.load("ToucHD-Force/", format="touchd", sensor="digit", obj_id=6, hand="r")

# All sensors
data = tlabel.load("ToucHD-Force/", format="touchd", sensor="all")

Install:

pip install --upgrade tlabel
pip install tlabel[touchd]  # with cv2 + scipy

Full Changelog: v0.5.3...v0.6.1

v0.5.3 Hot Fix #2

v0.5.3 — Hot Fix #2

Bug Fix

• 修复 generate_panel_html 的 NameError：renderDescribe() 中 JS 对象字面量 { i18n: ..., data: ... } 未在 Python f-string 中转义，导致 NameError: name 'i18n' is not defined

包含v0.5.2的修复

• 修复PyPI打包遗漏模块（adapters/batch/converters/core/demo_data等）

包含v0.5.1的修复

• i18n补全：雷达图22维+统计表格中英文切换
• 深色模式：统计页数字颜色修复
• Episode保存反馈：4秒+详细提示+防重复点击
• QUICKSTART.md全面更新

v0.5.1 Bug Fix Release

Bug Fixes

i18n补全

• 雷达图维度标签（22维特征）使用i18n国际化
• 统计摘要表格的行名和列名支持中英文切换

深色模式数字不可读

• 统计页和质量页在暗色模式下数字颜色修复
• 切换暗色模式后自动重新渲染统计数据

Episode语义标注保存反馈优化

• 反馈持续时间从2秒延长至4秒
• 反馈信息更详细
• 保存按钮点击后短暂禁用，防止重复点击

Documentation

• QUICKSTART.md 全面更新，使用 v0.5.0+ API