Benjamin_Boenisch/breakpilot-core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: disable oneDNN and support PaddleOCR 3.x format
Some checks failed
Deploy to Coolify / deploy (push) Failing after 2s
Details

Browse Source 
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Benjamin Admin
2026-03-13 18:54:28 +01:00
parent 52618a0630
commit 2ac6559291
1 changed files with 54 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
paddleocr-service/main.py
View File

@@ -5,6 +5,11 @@ 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["PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK"] = "1"
import numpy as np
from fastapi import FastAPI, File, Header, HTTPException, UploadFile
from PIL import Image
@@ -30,12 +35,10 @@ def _load_model():
logger.info("Import done. Loading PaddleOCR model...")
# Try multiple init strategies for different PaddleOCR versions
inits = [
# PaddleOCR 3.x (no show_log)
# PaddleOCR 3.x — use_textline_orientation replaces use_angle_cls
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 3.x with show_log
dict(lang="en", ocr_version="PP-OCRv5", use_angle_cls=True, show_log=False),
# PaddleOCR 2.8+ (latin)
dict(lang="latin", use_angle_cls=True, show_log=False),
# PaddleOCR 2.8+ (en, no version)
dict(lang="en", use_angle_cls=True, show_log=False),
]
@@ -94,26 +97,55 @@ async def ocr(
logger.error(f"OCR failed: {e}", exc_info=True)
raise HTTPException(status_code=500, detail=f"OCR processing failed: {e}")
if not result or not result[0]:
if not result:
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 = []
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)
x_max = max(p[0] for p in box)
y_max = max(p[1] for p in box)
words.append(
{
"text": text.strip(),
"left": int(x_min),
"top": int(y_min),
"width": int(x_max - x_min),
"height": int(y_max - y_min),
"conf": round(conf * 100, 1),
}
)
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)]
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)
x_max = max(p[0] for p in box)
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),
})
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,