[Bug] PaddleOCR v3.3.0 + PaddlePaddle v3.0.0 fails in Docker (ARM64 & x86_64)

[woojae05]

问题描述 Issue Description

🐛 PaddleOCR 3.3.0 + PaddlePaddle 3.0.0 fails in Docker environments (ARM64 & x86_64)

👋 Issue Summary

I'm experiencing critical compatibility issues when running PaddleOCR 3.3.0 with PaddlePaddle 3.0.0 in Docker containerized environments. The issues occur on both ARM64 (Apple Silicon) and x86_64 (AWS Lambda) architectures.

---

🔴 Problem Details

ARM64 Environment (Apple Silicon M-series)

Symptom: Segmentation Fault during model initialization

FatalError: `Segmentation fault` is detected by the operating system.
[SignalInfo: *** SIGSEGV (@0x0) received by PID 1 (TID 0xffff612341a0) from PID 0 ***]

Steps to Reproduce:

1. Build Docker image with PaddlePaddle 3.0.0 + PaddleOCR 3.3.0
2. Run basic initialization:

   from paddleocr import PaddleOCRocr = PaddleOCR(lang="en")  # ❌ Segfault here
   

---

x86_64 Environment (AWS Lambda Base Image)

Symptom: Library loading failure

❌ Error: name 'libpaddle' is not defined

Steps to Reproduce:

1. Use public.ecr.aws/lambda/python:3.11 base image
2. Install system dependencies (mesa-libGL, libXext, etc.)
3. Install PaddlePaddle 3.0.0 + PaddleOCR 3.3.0
4. Import succeeds, but initialization fails with libpaddle error

---

🔧 Environment Details

Item	Value
PaddleOCR Version	3.3.0
PaddlePaddle Version	3.0.0
Python Version	3.11
Base Images	python:3.11-slim (ARM64), public.ecr.aws/lambda/python:3.11 (x86_64)

---

🧪 Attempted Solutions

✅ Tried (No Effect)

1. Environment Variables:

   FLAGS_use_mkldnn=0
   FLAGS_use_xdnn=0
   OMP_NUM_THREADS=1
   OPENBLAS_NUM_THREADS=1
   MKL_NUM_THREADS=1
   

2. NumPy Version Pinning:

   pip install 'numpy<2.0.0'
   

3. PaddlePaddle Version Downgrade:

   • 3.0.0b2 (beta) → Same segfault
   • 2.6.1 → API compatibility issues (set_optimization_level missing)

4. Lightweight Models:

   • Switched to PP-OCRv4_mobile → Failed due to API parameter changes

---

📋 Reproducible Dockerfiles

ARM64 (Apple Silicon)

FROM python:3.11-slim
ENV FLAGS_use_mkldnn=0 

FLAGS_use_xdnn=0 

OMP_NUM_THREADS=1 

OPENBLAS_NUM_THREADS=1 

MKL_NUM_THREADS=1
RUN apt-get update && apt-get install -y 

libgl1 libglib2.0-0 libsm6 libxext6 

libxrender-dev libgomp1 

&& rm -rf /var/lib/apt/lists/*
RUN pip install --no-cache-dir 

paddlepaddle==3.0.0 

paddleocr==3.3.0 

'numpy<2.0.0' 

opencv-python-headless 

Pillow
WORKDIR /app

COPY test_paddleocr.py .

CMD ["python", "test_paddleocr.py"]

Build & Run:

docker build -t paddleocr-test .
docker run --rm paddleocr-test

---

x86_64 (AWS Lambda)

FROM public.ecr.aws/lambda/python:3.11
RUN yum update -y && 

yum install -y mesa-libGL libXext libSM libXrender libgomp && 

yum clean all
ENV FLAGS_use_mkldnn=0 

FLAGS_use_xdnn=0 

OMP_NUM_THREADS=1
RUN pip install --no-cache-dir 

paddlepaddle==3.0.0 

paddleocr==3.3.0 

'numpy<2.0.0' 

opencv-python-headless 

Pillow
WORKDIR /var/task

COPY test_paddleocr.py .

CMD ["python", "test_paddleocr.py"]

Build & Run:

docker buildx build --platform linux/amd64 -t paddle-lambda:test .
docker run --rm --platform linux/amd64 paddle-lambda:test

---

🧪 Test Script

from paddleocr import PaddleOCR
import numpy as np
import os
Environment optimization
os.environ["FLAGS_use_mkldnn"] = "0"

os.environ["FLAGS_use_xdnn"] = "0"

os.environ["GOTOBLAS_NUM_THREADS"] = "1"

os.environ["VECLIB_MAXIMUM_THREADS"] = "1"
print("✅ PaddleOCR import OK")
try:

print("Initializing PaddleOCR...")

ocr = PaddleOCR(lang="en")  # ❌ Fails here

print("✅ PaddleOCR init OK")
# Test prediction
img = np.ones((128, 512, 3), dtype=np.uint8) * 255
res = ocr.predict(img)
print("✅ Prediction OK")

except Exception as e:

print(f"❌ Error: {e}")

import traceback

traceback.print_exc()

---

🎯 Expected Behavior

PaddleOCR should initialize successfully in Docker containers on both ARM64 and x86_64 architectures without segmentation faults or library loading errors.

---

版本&环境信息 Version & Environment Information

🔧 Environment Details

Item	Value
PaddleOCR Version	3.3.0
PaddlePaddle Version	3.0.0
Python Version	3.11
Base Images	python:3.11-slim (ARM64), public.ecr.aws/lambda/python:3.11 (x86_64)