Detect windows doors soil / downspout etc

1

I have a working python script that detects faces cars etc streaming from a video source.
I’m learning CVAT so I can create my own .pt files.

The goal is to be able to take a video / pictures of a house and determine the following
How many door windows lengths of guttered downspouts waste pipes. Also internal and external / 45% bends etc.
Once collated produce a list of items.
I have used CVAT with images from googles street view but the images are too far from the house and low res so poor results .

  1. Are there any existing datasets of uk houses or plastic building products ?
  2. The training examples I have watched show drawing boxes around objects. Fine for doors and windows. For gutters do I need to draw exactly around the gutter or include the roof line.
  3. Soil pipes usually have a T junction into the house or via a bend. Do I classify as different types of complete pipework ( eg 0: has direct from house into T junction top and bottom pipes with 2 45 deg bends 1: elbow from house into T junction. Top bottom & bottom pipe) Or do I train on each separate component.
  4. Same with down spouts. From gutter outlet 2 45deg bends pipe to floor.

Any help or suggestions so I am not reinventing the wheel and I’m not going down the wrong image training route.

2

Hi guys,
I have gone through the process on installing CVAT on my local machine and trained on small set of house images with gutters, downspouts & waste pipes. I then exported to Yolo1.1 ( the only Yolo option in CVAT) which gave me a folder with a text file for each image. eg

2 0.199117 0.174856 0.395400 0.045975
2 0.670996 0.250719 0.230192 0.114938
1 0.410612 0.373634 0.058242 0.747269
0 0.530154 0.380531 0.082758 0.761062

I followed a tutorial that gave me the following python code and produced a train folder(see image) and a weights folder
with best.pt and last.pt.

image
image1056×651 153 KB

code

from ultralytics import YOLO

# Load a model
model = YOLO("yolov8n.yaml") 
model.train(data="config.yaml", epochs=10)

config.yaml

path: D:/techy/Vision/gutters/train
train: images/train
val : images/train

names:
  0: soilPipe
  1: wastePipe
  2: downSpout

I have a python that works with mp4 files and detects other objects when I use ‘yolov8x-oiv7.pt’

import cv2
from ultralytics import YOLO

def process_frame(frame, model):
    results = model(frame)
    if results:
        for result in results:
            if hasattr(result, 'boxes') and len(result.boxes) > 0:
                for box in result.boxes:
                    x1, y1, x2, y2 = box.xyxy[0].cpu().numpy()
                    conf = box.conf.cpu().numpy()[0]
                    cls_id = box.cls.cpu().numpy()[0]
                    label = f"{result.names[int(cls_id)]}: {conf:.2f}"

                    (text_width, text_height), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
                    cv2.rectangle(frame, (int(x1), int(y1) - 10 - text_height), (int(x1) + text_width, int(y1)), (255, 255, 255), cv2.FILLED)
                    cv2.putText(frame, label, (int(x1), int(y1) - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1)
                    cv2.rectangle(frame, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2)
    return frame

def main():
    #model = YOLO('best.pt')  # Load your model
    model = YOLO('yolov8x-oiv7.pt')
    video_path = 'soilpipe4.mp4'  # Path to your video file
    cap = cv2.VideoCapture(video_path)  # Open the video file

    if not cap.isOpened():
        print("Error: Failed to open video file.")
        return

    while True:
        ret, frame = cap.read()
        if not ret:
            print("End of video file reached.")
            break

        frame = process_frame(frame, model)

        # Display the processed frame
        cv2.imshow("YOLOv8 Object Detection - Video", frame)
        if cv2.waitKey(1) & 0xFF == ord('q'):  # Press 'q' to quit
            break

    cap.release()
    cv2.destroyAllWindows()

if __name__ == "__main__":
    main()

When I use best.pt I get no results. I changed the script to loop through the original images folders but got no detections.

I would presume that I would get something back.

Two thoughts.

  1. I have missed a step somewhere.
  2. The best.pt file is a different version that ( from ultralytics import YOLO).
    I can’t find a python script that converts from version Yolo 1.1 to 8.

Be gentle with me and answer at idiot level please.