⚔️ Longsword-Spatial-Physics-100 (BETA)

STATUS: IN PRODUCTION (Releasing Q3 2026) > This dataset is currently being captured and annotated. We are releasing this schema card to align with computer vision and robotics labs on data structure requirements before the final telemetry files go live.

📬 WANT ACTIVE BETA ACCESS? 👉 Click Here to Join the Waitlist

---

Dataset Description

Longsword-spatial-physics-100 is a high-velocity, biomechanical motion dataset specifically designed to stress-test human pose estimation and thin-object tracking models under conditions of extreme non-rigid occlusion.

Standard foundational vision models consistently collapse when human joints are hidden beneath bulky, uniform protective gear (HEMA fencing jackets) and when tracking objects (steel blades) that drop below sub-pixel resolution at high velocities (up to 80mph). This dataset acts as an edge-case alignment framework for bipedal locomotion, dynamic weight distribution, and high-speed trajectory tracking.

Key Specifications:

• Total Clips: 100 hyper-trimmed action sequence snippets.
• Duration: 3.0 – 5.0 seconds per clip.
• Framerate: 120 FPS / 240 FPS (Markerless raw RGB).
• Views: Synchronized Multi-View (Profile and Frontal camera arrays via Google Pixel hardware).
• Weapons Included: Historical Steel Longsword (Federschwert).

Intended Use Cases:

• Robotics (Bipedal Stabilization): Training reinforcement learning control policies to analyze high-torque, non-linear human weight shifts and rapid balance recovery.

• Industrial Computer Vision: Generalizing thin-object and hand-tracking tracking models to hazardous environments where workers wear bulky safety gear and handle fast-moving equipment.

• Generative Physics Alignment: Fine-tuning video models to understand realistic structural inertia, friction in a weapon bind, and real-world collision boundaries.

Contact & Collaborations:

Created by a specialized HEMA practitioner and data architecture enthusiast. For early enterprise licensing inquiries or customized physical routine captures, please join the waitlist or drop an issue in this repository.

---

Planned Data Schema (metadata.jsonl)

Every video clip maps to a corresponding row in a JSON Lines metadata file. Below is the active schema we are utilizing. If your lab requires additional keypoints (e.g., specific crossguard trajectory mapping), please request it via the waitlist form.

{
  "clip_id": "hema_ls_001",
  "meta": {
    "weapon": "Longsword",
    "source_text": "Joachim Meyer (1570)",
    "capture_fps": 120
  },
  "time_stamps": {
    "start_frame": 120,
    "blade_contact_frame": 165,
    "recovery_end_frame": 210
  },
  "biomechanics": {
    "initial_guard": "Right Vom Tag",
    "ending_guard": "Left Ochs",
    "footwork_type": "Passing step offline",
    "strike_trajectory": "Diagonal Oberhau",
    "edge_alignment": "True edge"
  },
  "computer_vision_hazards": {
    "occlusion_rating": "High (Crossed arms, bulky torso jacket)",
    "motion_blur_expected": true
  }
}