app.py

import cv2 
from ultralytics import YOLO ## for Yolov8
import matplotlib.pyplot as plt
import gradio as gr
import numpy as np
import pickle

# function which is returning the number of object detected 
def number_object_detected(image):
    
    custom_model = YOLO('runs/detect/train4/weights/best.pt') # custome yolo model path
    results = custom_model(image,verbose= False)

    dic = results[0].names
    classes = results[0].boxes.cls.cpu().numpy()
    probability = results[0].boxes.conf

    class_count = {}
    unique_elements, counts = np.unique(classes, return_counts=True)
    for e , count in zip(unique_elements,counts):
        a = dic[e]
        class_count[a] = count
    print(class_count)
    return (class_count,results )


def car_detection_and_Cropping(image_path):
    simple_yolo = YOLO('yolov8m.pt')
    r = simple_yolo(image_path,verbose = False)
    

    names = r[0].names
    boxes = r[0].boxes.xyxy.cpu().numpy().astype(int)
    classes = set(r[0].boxes.cls.cpu().numpy())
    classes2 =  [names[i] for i in classes]

    # checking if the detected object is the car or not 
    # if it is car then crop if not then pass the image as it is
    if boxes.size != 0 and 'car' in classes2:
      
      area = []
      for x1, y1, x2, y2 in boxes:
          area.append((x2 - x1) * (y2 - y1))
      max_index, max_a = max(enumerate(area), key=lambda x: x[1])
      

      # Load the image using OpenCV
      image = cv2.imread(image_path)
      
      # Crop the image
      crop_image = image[boxes[max_index][1]:boxes[max_index][3], boxes[max_index][0]:boxes[max_index][2]]

      # passing the crop image to the detection model

      class_c ,result = number_object_detected(crop_image)
    else:
      class_c ,result= number_object_detected(image_path)
    return class_c ,result


severity_points = {
    'scratch': 1,
    'dent': 2,
    'rust': 2,
    'paint-damage': 2,
    'crack':2
}

def calculate_condition_score(detections):
    total_score = 0
    for detection, count in detections.items():
        if detection in severity_points:
            total_score += severity_points[detection] * count
    return total_score

def normalize_score(score, max_score):
    return (score / max_score) * 10


## this function will take the image url and call all the related functions
def estimate_condition(detections):
    print("Detedtion list",detections)
    max_possible_score = sum(severity_points.values())  # Assuming all types of damage detected
    score = calculate_condition_score(detections)
    normalized_score = normalize_score(score, max_possible_score)
    
    if normalized_score <= 2:  # If score is low, condition is Excellent
        print("Condition Excellent")
        return "Excellent"
    elif (normalized_score >2 and normalized_score <=7):  # If score is moderately low, condition is Good
        print("Condition Good")
        return "Good"
    elif (normalized_score >7 and normalized_score <15):  # If score is moderate, condition is Fair
        print("Condition Fair")
        return "Fair"
    elif (normalized_score >15 and normalized_score<=20):  # If score is moderately high, condition is Poor
        print("Condition Poor")
        return "Poor"
    else:  # If score is high, condition is Very Poor
        print("Condition Very Poor")
        return "Very Poor"




## loading the model
def process_data(files):
    print(files)
    file_names = [f[0] for f in files]
    image_r = []
    print('fileName',file_names)
    
    damage_dic = {}

    for f in file_names:
        print('image is ',f)
        damage, result = car_detection_and_Cropping(f)
        for r in result:
            im_array = r.plot(pil = True)  # plot a BGR numpy array of predictions
            array = im_array[..., ::-1]  # Convert BGR to RGB PIL image
            image_r.append(array)
        for key in damage.keys():
                      if key in damage_dic:
                        damage_dic[key] += damage[key]
                      else:
                          damage_dic[key] = damage[key]
    condition = estimate_condition(damage_dic)


    return (condition,image_r)

interface = gr.Interface(fn=process_data, inputs=gr.Gallery(label='Upload Image of Car',type= 'filepath'), 
                         outputs=[gr.Textbox(label="Number of Objects detected "),gr.Gallery(label='output',type='pil')], title=" 🚘Car Scratch and Dent Detection")
interface.launch()