feat: camera-free 17-keypoint pose training (10 sensor signals)

Browse Source

Multi-modal pipeline using PIR, BME280, reed switch, vibration,
RSSI triangulation, subcarrier asymmetry — no camera needed.

Phases: multi-modal collection → weak label generation → enhanced
contrastive → 5-keypoint pose proxy → 17-keypoint interpolation
→ self-refinement (3 rounds) → LoRA + TurboQuant + EWC

Validated: 2,360 frames, 100% presence, 0 skeleton violations,
82.8 KB model (8 KB at 4-bit), 114.8s training

Co-Authored-By: claude-flow <ruv@ruv.net>

...

This commit is contained in:

ruv

2026-04-02 23:05:07 -04:00

parent ccc543c0e7

commit ba82fcfc37

2 changed files with 2621 additions and 0 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

scripts/train-camera-free.js

+2489

File diff suppressed because it is too large Load Diff