Poulpe 13323f2edf fix: 05_inference — kill stale demo.py + background poll exit viser + offload_to_cpu from yaml
- kill_stale_demo_py() before each segment to prevent GPU contention from orphan processes
- Remote script runs demo.py in background via nohup, polls for PLY file every 30s, kills viser server once PLY written — prevents indefinite SSH block on viser listener
- offload_to_cpu now read from thresholds.yaml[inference] (default false for 24GB VRAM)
- timeout reads inference_timeout_s from yaml (already 10800s)
- min_frames guard included (from fix/05-inference-min-frames-timeout)

Root cause: demo.py starts viser server after writing PLY; SSH timed out → orphan; two orphans competed for GPU with offload_to_cpu → pure CPU inference = 6h+ for 493 frames
2026-05-13 16:41:18 +00:00

cosma-qc

COSMA post-acquisition QC pipeline — reconstruction photogrammétrique par GoPro (lingbot-map), queue de jobs distribués, dashboard web pour suivi terrain le jour même.

Objectif

Après une acquisition AUV (2 GoPros × 2-3 AUVs × heures d'enregistrement), savoir rapidement si la couverture est complète avant de replier la mission — sans attendre les 30 jours du traitement photogrammétrique complet.

Pipeline

SSD plugged ─┐
             ├─▶ Ingestion ─▶ Frame extraction (per GoPro × segment)
             │                        │
             │                        ▼
             │                  Job queue (SQLite)
             │                        │
             │         ┌──────────────┼──────────────┐
             ▼         ▼              ▼              ▼
         Dashboard  Worker .87    Worker .84    (scalable)
         (FastAPI)   (3060)       (3090)
             │         │              │
             │         └─▶ PLY ◀──────┘
             │              │
             │              ▼
             └──────── ICP stitch (Open3D) ─▶ viser viewer

Stack

  • Backend : FastAPI + SQLite
  • Frontend : HTMX (UI réactive sans build JS)
  • Queue : table SQLite + workers SSH-triggered
  • Monitoring : polling nvidia-smi sur .87 / .84, df pour disque
  • Reconstruction : lingbot-map (GCT-Stream windowed)
  • Stitch : Open3D ICP

Déploiement

  • Service sur .82 (stable, Caddy pour URL propre)
  • Workers : SSH vers .87 (3060 12 GB) et .84 (3090 24 GB)

État

Scaffold en cours.

Description
COSMA post-acquisition QC pipeline: per-GoPro lingbot-map reconstruction, job queue, web dashboard
Readme 6.3 MiB
Languages
Python 76.3%
HTML 11.6%
CSS 11.5%
Dockerfile 0.6%