fix: downgrade to PaddleOCR 2.x for CPU stability
Some checks failed
Deploy to Coolify / deploy (push) Failing after 2s

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-03-13 19:13:41 +01:00
parent 6adf1fe1eb
commit 520a0f401c
2 changed files with 35 additions and 89 deletions

View File

@@ -1,23 +1,10 @@
"""PaddleOCR Remote Service — PP-OCRv5 Latin auf x86_64."""
"""PaddleOCR Remote Service — PP-OCRv4 on x86_64 (CPU)."""
import io
import logging
import os
import threading
# Disable oneDNN/MKLDNN before importing paddle — avoids
# ConvertPirAttribute2RuntimeAttribute errors on PaddlePaddle 3.x
os.environ["FLAGS_use_mkldnn"] = "0"
os.environ["FLAGS_use_onednn"] = "0"
os.environ["PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK"] = "1"
# Must set paddle flags before import
try:
import paddle
paddle.set_flags({"FLAGS_use_mkldnn": False})
except Exception:
pass
import numpy as np
from fastapi import FastAPI, File, Header, HTTPException, UploadFile
from PIL import Image
@@ -40,31 +27,22 @@ def _load_model():
logger.info("Importing paddleocr...")
from paddleocr import PaddleOCR
logger.info("Import done. Loading PaddleOCR model...")
# Try multiple init strategies for different PaddleOCR versions
inits = [
# PaddleOCR 3.x — disable MKLDNN via enable_mkldnn=False
dict(lang="en", ocr_version="PP-OCRv5", use_textline_orientation=True, enable_mkldnn=False),
# PaddleOCR 3.x without enable_mkldnn param
dict(lang="en", ocr_version="PP-OCRv5", use_textline_orientation=True),
# PaddleOCR 3.x with deprecated param
dict(lang="en", ocr_version="PP-OCRv5", use_angle_cls=True),
# PaddleOCR 2.8+ (en, no version)
dict(lang="en", use_angle_cls=True, show_log=False),
]
for i, kwargs in enumerate(inits):
try:
_engine = PaddleOCR(**kwargs)
logger.info(f"PaddleOCR init succeeded with strategy {i}: {kwargs}")
break
except Exception as e:
logger.info(f"PaddleOCR init strategy {i} failed: {e}")
else:
raise RuntimeError("All PaddleOCR init strategies failed")
logger.info("Loading PaddleOCR model (PP-OCRv4, lang=en)...")
_engine = PaddleOCR(
lang="en",
use_angle_cls=True,
show_log=False,
enable_mkldnn=False,
use_gpu=False,
)
logger.info("PaddleOCR model loaded — running warmup...")
# Warmup with tiny image to trigger any lazy init
dummy = np.ones((30, 100, 3), dtype=np.uint8) * 255
_engine.ocr(dummy)
_ready = True
logger.info("PaddleOCR model loaded successfully — ready to serve")
logger.info("PaddleOCR ready to serve")
except Exception as e:
logger.error(f"Failed to load PaddleOCR model: {e}")
logger.error(f"Failed to load PaddleOCR: {e}", exc_info=True)
@app.on_event("startup")
@@ -72,15 +50,14 @@ def startup_load_model():
"""Start model loading in background so health check passes immediately."""
global _loading
_loading = True
thread = threading.Thread(target=_load_model, daemon=True)
thread.start()
threading.Thread(target=_load_model, daemon=True).start()
logger.info("Model loading started in background thread")
@app.get("/health")
def health():
if _ready:
return {"status": "ok", "model": "PP-OCRv5-latin"}
return {"status": "ok", "model": "PP-OCRv4"}
if _loading:
return {"status": "loading"}
return {"status": "error"}
@@ -105,43 +82,13 @@ async def ocr(
result = _engine.ocr(img_np)
except Exception as e:
logger.error(f"OCR failed: {e}", exc_info=True)
raise HTTPException(status_code=500, detail=f"OCR processing failed: {e}")
raise HTTPException(status_code=500, detail=f"OCR failed: {e}")
if not result:
if not result or not result[0]:
return {"words": [], "image_width": img_np.shape[1], "image_height": img_np.shape[0]}
# PaddleOCR 2.x returns: [[line, ...]] where line = [box, (text, conf)]
# PaddleOCR 3.x returns: [{'text': ..., 'boxes': [...], 'rec_scores': ...}] or similar
words = []
try:
lines = result[0] if isinstance(result, list) and result else result
if not lines:
return {"words": [], "image_width": img_np.shape[1], "image_height": img_np.shape[0]}
for line in lines:
if isinstance(line, dict):
# PaddleOCR 3.x dict format
text = str(line.get("text", line.get("rec_text", ""))).strip()
conf = float(line.get("score", line.get("rec_score", 0)))
box = line.get("boxes", line.get("dt_polys", []))
if not text or not box:
continue
# box might be [[x1,y1],[x2,y2],[x3,y3],[x4,y4]] or flat
if isinstance(box[0], (list, tuple)):
x_min = min(p[0] for p in box)
y_min = min(p[1] for p in box)
x_max = max(p[0] for p in box)
y_max = max(p[1] for p in box)
else:
x_min, y_min, x_max, y_max = box[0], box[1], box[2], box[3]
words.append({
"text": text,
"left": int(x_min), "top": int(y_min),
"width": int(x_max - x_min), "height": int(y_max - y_min),
"conf": round(conf * 100 if conf <= 1 else conf, 1),
})
elif isinstance(line, (list, tuple)) and len(line) == 2:
# PaddleOCR 2.x format: [box, (text, conf)]
for line in result[0]:
box, (text, conf) = line[0], line[1]
x_min = min(p[0] for p in box)
y_min = min(p[1] for p in box)
@@ -149,13 +96,12 @@ async def ocr(
y_max = max(p[1] for p in box)
words.append({
"text": str(text).strip(),
"left": int(x_min), "top": int(y_min),
"width": int(x_max - x_min), "height": int(y_max - y_min),
"conf": round(float(conf) * 100 if conf <= 1 else float(conf), 1),
"left": int(x_min),
"top": int(y_min),
"width": int(x_max - x_min),
"height": int(y_max - y_min),
"conf": round(float(conf) * 100, 1),
})
except Exception as e:
logger.error(f"Failed to parse OCR result: {e}. Raw: {str(result)[:500]}", exc_info=True)
raise HTTPException(status_code=500, detail=f"OCR result parsing failed: {e}")
return {
"words": words,

View File

@@ -1,5 +1,5 @@
paddlepaddle>=3.0.0
paddleocr>=2.9.0
paddlepaddle>=2.6.0,<3.0.0
paddleocr>=2.7.0,<3.0.0
fastapi>=0.110.0
uvicorn>=0.25.0
python-multipart>=0.0.6