updated roboflow model to improve accuracy

app.py

@@ -39,6 +39,8 @@ from PIL import Image
 import os
 from sklearn.model_selection import train_test_split
 from sklearn.preprocessing import MinMaxScaler
+from inference_sdk import InferenceHTTPClient
+
 # Initialize Flask app
 app = Flask(__name__)
 
@@ -138,88 +140,133 @@ def fetchImage():
         scene=image, detections=detections_damage
     )
 
-    # Create a temporary directory to save outputs
-    temp_dir = tempfile.mkdtemp()
+    # temp_dir = tempfile.mkdtemp()
 
     # Define a repair cost dictionary (per part)
-    repair_cost_dict = {
-        "wheel": 100,  # Base cost for wheel
-        "door": 200,  # Base cost for door
-        "hood": 300,  # Base cost for hood
-        "front_bumper": 250,  # Base cost for bumper
-        "trunk": 200,
-        "front_glass": 150,
-        "back_left_door": 200,
-        "left_mirror": 20,
-        "back_glass": 150,
+    repair_costs = {
+        "Car-Damage-Detection-1KxY": 1000,  # General damage assessment cost
+        "Bodypanel-Dent": 200,
+        "Front-Windscreen-Damage": 400,
+        "Headlight-Damage": 250,
+        "Rear-windscreen-Damage": 350,
+        "RunningBoard-Dent": 150,
+        "Sidemirror-Damage": 180,
+        "Signlight-Damage": 120,
+        "Taillight-Damage": 220,
+        "back-bumper": 500,
+        "back-glass": 400,
+        "bonnet-dent": 300,
+        "boot-dent": 350,
+        "broken_lamp": 100,
+        "crack": 250,
+        "damaged-door": 600,
+        "damaged-front-bumper": 550,
+        "damaged-head-light": 270,
+        "damaged-hood": 500,
+        "damaged-rear-bumper": 520,
+        "damaged-rear-window": 380,
+        "damaged-tail-light": 230,
+        "damaged-trunk": 600,
+        "damaged-window": 280,
+        "damaged-windscreen": 450,
+        "dent": 200,
+        "dent-or-scratch": 180,
+        "door": 700,
+        "doorouter-dent": 250,
+        "fender-dent": 220,
+        "flat_tire": 100,
+        "front-bumper": 500,
+        "front-bumper-dent": 450,
+        "front-glass": 400,
+        "headlight": 250,
+        "hood": 500,
+        "mirror": 180,
+        "pillar-dent": 220,
+        "quaterpanel-dent": 270,
+        "rear-bumper-dent": 480,
+        "roof-dent": 400,
+        "scratch": 150,
+        "shattered_glass": 500,
+        "taillight": 220,
+        "trunk": 600,
+        "wheel": 250,
+        "window": 300,
     }
 
-    # Initialize total cost
     total_cost = 0
 
-    # Ensure coordinate processing is done in chunks of 4
-    coordinates = list(map(int, detections_damage.xyxy.flatten()))
-    num_damages = (
-        len(coordinates) // 4
-    )  # Each damage has 4 coordinates (x1, y1, x2, y2)
+    # coordinates = list(map(int, detections_damage.xyxy.flatten()))
+    # num_damages = (
+    #     len(coordinates) // 4
+    # )  # Each damage has 4 coordinates (x1, y1, x2, y2)
 
     # Iterate through damages
-    for i in range(num_damages):
-        x1, y1, x2, y2 = coordinates[i * 4: (i + 1) * 4]
-
-        # Ensure the coordinates are within image bounds
-        x1, y1 = max(0, x1), max(0, y1)
-        x2, y2 = min(image.shape[1], x2), min(image.shape[0], y2)
-
-        # Crop the damaged region
-        cropped_damage = image[y1:y2, x1:x2]
-
-        # Check if the cropped region is valid
-        if cropped_damage.size == 0:
-            print(f"Skipping empty crop for damage region {i + 1}")
-            continue
-
-        # Save the cropped damaged area
-        damage_image_path = os.path.join(temp_dir, f"damage_image_{i}.png")
-        cv2.imwrite(damage_image_path, cropped_damage)
-
-        # Run the parts detection model on the cropped damage
-        result_parts = model_parts.predict(
-            damage_image_path, confidence=15).json()
-        detections_parts = sv.Detections.from_inference(result_parts)
-
-        # Calculate repair cost for each detected part
-        for part in result_parts["predictions"]:
-            part_name = part["class"]
-            damage_area = part["width"] * part["height"]
-            cropped_area = (x2 - x1) * (y2 - y1)
-            damage_percentage = (damage_area / cropped_area) * 100
-
-            # Lookup cost and add to total
-            base_cost = repair_cost_dict.get(
-                part_name, 0
-            )  # Default to 0 if part not in dict
-            repair_cost = (damage_percentage / 100) * 10 * base_cost
-            total_cost += round(repair_cost, ndigits=1)
-
-            print(
-                f"Damage {i + 1} - {part_name}: {damage_percentage:.2f}% damaged, Cost: ${repair_cost:.2f}"
-            )
-
-        # Annotate and save the result
-        part_annotator = sv.LabelAnnotator()
-        annotated_parts_image = part_annotator.annotate(
-            scene=cropped_damage, detections=detections_parts
-        )
-        annotated_parts_path = os.path.join(
-            temp_dir, f"annotated_parts_{i}.png")
-        cv2.imwrite(annotated_parts_path, annotated_parts_image)
-
-    # Save the overall annotated image
-    annotated_image_path = os.path.join(temp_dir, "annotated_image_damage.png")
-    cv2.imwrite(annotated_image_path, annotated_image_damage)
-
-    # Return the total cost in the specified format
+    # for i in range(num_damages):
+    #     x1, y1, x2, y2 = coordinates[i * 4: (i + 1) * 4]
+
+    #     # Ensure the coordinates are within image bounds
+    #     x1, y1 = max(0, x1), max(0, y1)
+    #     x2, y2 = min(image.shape[1], x2), min(image.shape[0], y2)
+
+    #     # Crop the damaged region
+    #     cropped_damage = image[y1:y2, x1:x2]
+
+    #     # Check if the cropped region is valid
+    #     if cropped_damage.size == 0:
+    #         print(f"Skipping empty crop for damage region {i + 1}")
+    #         continue
+
+    #     # Save the cropped damaged area
+    #     damage_image_path = os.path.join(temp_dir, f"damage_image_{i}.png")
+    #     cv2.imwrite(damage_image_path, cropped_damage)
+
+    #     # Run the parts detection model on the cropped damage
+    #     result_parts = model_parts.predict(
+    #         damage_image_path, confidence=15).json()
+    #     detections_parts = sv.Detections.from_inference(result_parts)
+
+    #     # Calculate repair cost for each detected part
+    #     for part in result_parts["predictions"]:
+    #         part_name = part["class"]
+    #         damage_area = part["width"] * part["height"]
+    #         cropped_area = (x2 - x1) * (y2 - y1)
+    #         damage_percentage = (damage_area / cropped_area) * 100
+
+    #         # Lookup cost and add to total
+    #         base_cost = repair_cost_dict.get(
+    #             part_name, 0
+    #         )  # Default to 0 if part not in dict
+    #         repair_cost = (damage_percentage / 100) * 10 * base_cost
+    #         total_cost += round(repair_cost, ndigits=1)
+
+    #         print(
+    #             f"Damage {i + 1} - {part_name}: {damage_percentage:.2f}% damaged, Cost: ${repair_cost:.2f}"
+    #         )
+
+    #     # Annotate and save the result
+    #     part_annotator = sv.LabelAnnotator()
+    #     annotated_parts_image = part_annotator.annotate(
+    #         scene=cropped_damage, detections=detections_parts
+    #     )
+    #     annotated_parts_path = os.path.join(
+    #         temp_dir, f"annotated_parts_{i}.png")
+    #     cv2.imwrite(annotated_parts_path, annotated_parts_image)
+
+    # # Save the overall annotated image
+    # annotated_image_path = os.path.join(temp_dir, "annotated_image_damage.png")
+    # cv2.imwrite(annotated_image_path, annotated_image_damage)
+
+    # # Return the total cost in the specified format
+
+    CLIENT = InferenceHTTPClient(
+        api_url="https://detect.roboflow.com",
+        api_key="LqD8Cs4OsoK8seO3CPkf"
+    )
+    result = CLIENT.infer(file_name, model_id="car-damage-detection-krsix/1")
+    labels = [item["class"] for item in result["predictions"]]
+    print(labels)
+    for class_ in labels:
+        total_cost += repair_costs.get(class_, 0)
     result = {"total_cost": total_cost}
     print(result)
 

requirements.txt

@@ -19,6 +19,7 @@ opencv-python
 requests
 PyMuPDF
 openai
+inference-sdk
 fpdf
 cloudinary
 PyPDF2