app.py

import streamlit as st
import tensorflow as tf
from PIL import ImageOps, Image
import os                        # To work with operation system commands
import cv2                       # To process images
import random                    # to generate random choices
import warnings                  # To avoid python warnings
import numpy as np               # To work with arrays
import pandas as pd              # To work with DataFrames
import seaborn as sns            # To visualizations
from tqdm import tqdm            # To represent progress bars
from ultralytics import YOLO     # To Create Yolo model
from termcolor import colored    # To colorfull outputs
import matplotlib.pyplot as plt  # To visualizations

#load weights
best_model_plate = YOLO('best.pt')
best_model_digits = YOLO('best2.pt')

def Detect_Plate(img):
    '''
    Predict on a car image and return result folder and predicted image path. 
    '''
    result1 = best_model_plate.predict([img], save=True, iou=0.7)
    
    # Location of saved predicted images
    result_path1 = result1[0].save_dir
    pred1_path = os.path.join(result_path1, os.listdir(result_path1)[-1])
    return result1, pred1_path

def Detect_Digits(img2) :
    '''
    Predict on a croped plate and return result folder and predicted image.
    '''
    result2 = best_model_digits.predict([img2], save=True, iou=0.7, show_conf=False)
    
    # Location of saved predicted images
    result_path = result2[0].save_dir
    pred2_path = os.path.join(result_path, os.listdir(result_path)[0])
    return result2, pred2_path

def Sort_Digits(result2) :
    '''
    Sort detected labels base on their X-cordinate,
    --- Sort from lef to right ---
    '''
    list_of_items = []
    for i in range(len(result2[0])) :
        # Class labels decoded
        class_label = classes_dict[str(int(result2[0][i].boxes.cls))]
        # X cordinate to find the situation on X-axis
        cord_x = int(result2[0][i].boxes.xyxy[0][0])
        list_of_items.append((cord_x, class_label))

    list_of_items.sort()
    list_of_items

    digits = []
    for digit in list_of_items :
        digits.append(digit[1])
    if len(digits) == 7 : digits.append('0')
    return digits


raw_plate = ('raw_plate.png')

def FINAL(img) :
    '''
    A pipeline for all parts of phase 3.
    start with a car image.
    result is digits and char on car plate.
    '''
    # Read car image ( STEP-1 )
    #img = cv2.imread(img)
    #img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    img = Image.open(img).convert('RGB')

    # First prediction -> Detect car-plate ( STEP-2 )
    result1, _ = Detect_Plate(img)

    # Plate Cordinates ( STEP-3 )
    pts = result1[0].boxes.xyxy.tolist()[0]

    # Crop plate
    img2 = img[round(pts[1]):round(pts[3]), round(pts[0]):round(pts[2])]

    # Resize plate to feed to second model ( STEP-4 )
    img2 = tf.image.resize(img2, (120, 70))

    # Second prediction -> Detect digits in plate
    result2, _ = Detect_Digits(img2)

    # Sort detected digits ( STEP-5 )
    digits = Sort_Digits(result2)

    Plot_Result(img, raw_plate, digits)

# set title
st.title('Persian Plates Digits Detection')

# set header
st.header('Please upload image')

# upload file
img = st.file_uploader('', type=['jpeg', 'jpg', 'png'])

# display image
if img is not None:
    image = Image.open(img).convert('RGB')
    st.image(image, use_column_width=True)


st.write(FINAL(img))