App always crashes when it comes to analysing pictures with EasyOCR

[loeerc]

Hallo everyone ,
I need your help. I am currently going through some pictures which I got from URL and just want to analyze them with easyocr bit everytime the model starts to analyze the app crahes and this is what I get:

Can please someone help me?

[Franky1]

Unfortunately this is not enough information to help.
I suspect that the app is crashing because it is hitting memory limits.
But i could be wrong.

[loeerc]

import os
import cv2
from PIL import Image
from PIL import ImageOps
import numpy as np
import re
import streamlit as st
import easyocr
import requests
import tempfile
import shutil

@st.cache
def load_model():
    return easyocr.Reader(['de'],gpu=False)
#reader = load_model()
def get_context_R(url_pages):
    reader = load_model()
    #os.chdir(path_pict) 
    
    #texts_detected={}
    #number_pictures=str(len(os.listdir('./')))
    #dirFiles=os.listdir('./')
    #dirFiles.sort(key=lambda f: int(str(f).split('_')[-1].split('.')[0]))
    
    display=st.empty()
    for url in url_pages:
       
        text="Analysing context of page: {page} / {number_pictures}".format(page=str(url).split('_page_')[-1].split('.jpg')[0],number_pictures=str(len(url_pages)))
        display.metric("Currently",f"{text}")
        
        response = requests.get(url, stream=True)
        response.raw.decode_content = True
        fp=tempfile.NamedTemporaryFile(delete=False,suffix='.jpg')
        name=fp.name
        shutil.copyfileobj(response.raw, fp)
        image_full = Image.open(fp)
        width, height = image_full.size
        fp.seek(0)
        croppedImage = image_full.crop((0, 0, width, height/3*2)).save(fp,'JPEG' ,quality=95)
            #image=fp
        img=cv2.imread(fp.name)
        list_text=reader.readtext(img, detail = 0)
        if 'KNALLER' in list_text or 'AKTION' in list_text:
           #print(str(filename).split('_')[-1].split('.')[0])
           texts_detected[str(url).split('_page_')[-1].split('.jpg')[0]]='KNALLER AKTION'
        elif int(str(url).split('_page_')[-1].split('.jpg')[0])==0:
            texts_detected[str(url).split('_page_')[-1].split('.jpg')[0]]='Deine Auswahl - Auch beim Preis '+reader.readtext(img, detail = 0)[2]+' '+reader.readtext(img, detail = 0)[3]
        else:
            texts_detected[str(url).split('_page_')[-1].split('.jpg')[0]]=' '.join(reader.readtext(img, detail = 0)[0:4])
        os.unlink(fp.name)
    load_model.clear()
    return texts_detected

Maybe this part of my script which gets a list of urls could help

[Franky1]

Which streamlit version do you use? Seems to be not the latest version?
I would use the latest version with this:

@st.cache_resource
def load_model():
    return easyocr.Reader(['de'], gpu=False)

[loeerc]

streamlit==1.21.0

[dataprofessor]

Hi @loeerc

As @Franky1 pointed out, I also think that the app may reach the memory limit. Have you tried running locally to see the memory usage.

Also, have you tried running with a fewer number of URLs as input to the app.

Best regards,
Chanin

[loeerc]

Thanks for your answers @dataprofessor & @Franky1 ,

I will just check if it runs locally fine and how much memory it consumes as well. Thanks for the suggestion.

[Franky1]

I think EasyOCR has some very heavy dependencies e.g. torch etc.
I quickly ran a docker container locally with:

pip install easyocr

and this:

import easyocr
model = easyocr.Reader(['de'], gpu=False)

These are my resources eaten up by the container:

• 350MB RAM
• 7.5GB Disk

[loeerc]

Yes, I think that’s the point even if I just take one url the app continues to crash