Dom/rpa_vision_v3
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

perf: RF-DETR sur GPU (cuda) — était sur CPU = 28s par étape
Some checks failed
security-audit / Bandit (scan statique) (push) Successful in 12s
Details
security-audit / pip-audit (CVE dépendances) (push) Successful in 10s
Details
security-audit / Scan secrets (grep) (push) Successful in 9s
Details
tests / Lint (ruff + black) (push) Successful in 14s
Details
tests / Tests unitaires (sans GPU) (push) Failing after 15s
Details
tests / Tests sécurité (critique) (push) Has been skipped
Details

Browse Source 
RF-DETR détecte 90+ éléments UI par screenshot. Sur CPU = 28s.
Sur GPU RTX 5070 = devrait être 1-3s.

CLIP auto-GPU déjà en place (vérifie 1.5 Go VRAM libre).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Dom
2026-04-21 18:54:19 +02:00
parent 7feef3b6a9
commit 7b50725bf8
1 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
visual_workflow_builder/backend/services/ui_detection_service.py
View File

@@ -108,10 +108,12 @@ def _load_rfdetr():
return _rfdetr_model
from rfdetr.detr import RFDETRMedium
print(f"[UI-DETR-1] Chargement du modèle...")
import torch
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"[UI-DETR-1] Chargement du modèle sur {device}...")
start = time.time()
_rfdetr_model = RFDETRMedium(pretrain_weights=MODEL_PATH, resolution=RESOLUTION)
print(f"[UI-DETR-1] Modèle chargé en {time.time() - start:.1f}s")
_rfdetr_model = RFDETRMedium(pretrain_weights=MODEL_PATH, resolution=RESOLUTION, device=device)
print(f"[UI-DETR-1] Modèle chargé en {time.time() - start:.1f}s ({device})")
return _rfdetr_model