Live Object Detection and Image Segmentation with YOLOv8

After pressing the “Ctrl+x”. you can exit the file. and going to the main path.

Step5: Installing the requirements.txt

pip install -r requirements.txt

Step6: Create the Python Script

In the terminal write the following script or we can say command.

touch main.py

After creating main.py open the vs code you use the command write in terminal,

code 

Step7: Writing the Python Script

import cv2
from ultralytics import YOLO
import supervision as sv

# Define the frame width and height for video capture
frame_width = 1280
frame_height = 720

def main():
    # Initialize video capture from default camera
    cap = cv2.VideoCapture(0)
    cap.set(cv2.CAP_PROP_FRAME_WIDTH, frame_width)
    cap.set(cv2.CAP_PROP_FRAME_HEIGHT, frame_height)

    # Load YOLOv8 model
    model = YOLO("yolov8l.pt")

    # Initialize box annotator for visualization
    box_annotator = sv.BoxAnnotator(
        thickness=2,
        text_thickness=2,
        text_scale=1
    )

    # Main loop for video processing
    while True:
        # Read frame from video capture
        ret, frame = cap.read()

        # Perform object detection using YOLOv8
        result = model(frame, agnostic_nms=True)[0]
        detections = sv.Detections.from_yolov8(result)

        # Prepare labels for detected objects
        labels = [
            f"{model.model.names[class_id]} {confidence:0.2f}"
            for _, confidence, class_id, _
            in detections
        ]

        # Annotate frame with bounding boxes and labels
        frame = box_annotator.annotate(
            scene=frame,
            detections=detections,
            labels=labels
        )

        # Display annotated frame
        cv2.imshow("yolov8", frame)

        # Check for quit key
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break

    # Release video capture
    cap.release()
    cv2.destroyAllWindows()

if __name__ == "__main__":
    main()

After running this command you can see the your camera is open and detecting part of you. like gender and background parts.

Step7: Create streamlit App

import cv2
import streamlit as st
from ultralytics import YOLO
import supervision as sv

# Define the frame width and height for video capture
frame_width = 1280
frame_height = 720

def main():
    # Set page title and header
    st.title("Live Object Detection with YOLOv8")

    # Button to start the camera
    start_camera = st.button("Start Camera")

    if start_camera:
        # Initialize video capture from default camera
        cap = cv2.VideoCapture(0)
        cap.set(cv2.CAP_PROP_FRAME_WIDTH, frame_width)
        cap.set(cv2.CAP_PROP_FRAME_HEIGHT, frame_height)

        # Load YOLOv8 model
        model = YOLO("yolov8l.pt")

        # Initialize box annotator for visualization
        box_annotator = sv.BoxAnnotator(
            thickness=2,
            text_thickness=2,
            text_scale=1
        )

        # Main loop for video processing
        while True:
            # Read frame from video capture
            ret, frame = cap.read()

            # Perform object detection using YOLOv8
            result = model(frame, agnostic_nms=True)[0]
            detections = sv.Detections.from_yolov8(result)

            # Prepare labels for detected objects
            labels = [
                f"{model.model.names[class_id]} {confidence:0.2f}"
                for _, confidence, class_id, _
                in detections
            ]

            # Annotate frame with bounding boxes and labels
            frame = box_annotator.annotate(
                scene=frame,
                detections=detections,
                labels=labels
            )

            # Display annotated frame
            st.image(frame, channels="BGR", use_column_width=True)

            # Check for quit key
            if cv2.waitKey(1) & 0xFF == ord('q'):
                break

        # Release video capture
        cap.release()

if __name__ == "__main__":
    main()

In this script, we are creating the streamlit application and creating the button so that after pressing the button your device camera is open and detecting the part in the frame.

Run this script using this command.

streamlit run app.py
# first create the app.py then paste the above code and run this script.

After running the above command suppose you got the reach-out error like,

then press this command,

sudo sysctl fs.inotify.max_user_watches=524288

After hitting the command you want to write your password by because we are using the sudo command sudo is god:)

Run the script again. and you can see the streamlit application.

Here we can create a successful live detection application in the next part we will see the segmentation part.

Steps For Annotation

Step1: Roboflow Setup

After sigin the “Create Project”. here you can create the project and annotation group.

Step2: Dataset Downloading

Here we consider the simple example but you want to use it on your problem statement so I’m using here duck dataset.

Go the this link and download the duck dataset.

Extract the folder there you can see the three folders: train, test and val.

Step3: Uploading the Dataset on roboflow

After creating the project in roboflow you can see this UI here you can upload your dataset so are uploading only train part images select the “select folder” option.

Then click the “save and continue” option as I mark in a red rectangle box

Step4: Add the Class Name

Then go to the class part on the left side of check the red box. and write the class name as duck, after clicking the green box.

Now our setup is complete and the next part like the annotation part also simple.

Step5: Start the annotation part

Go to the annotation option I marked in the red box and then click the start the annoataion part as I marked in the green box.

Click the first image you can see this UI. After seeing this click the manual annotation option.

Then add your email id or your teammate’s name so you can assign the task.

Click the first image you can see this UI. here click the red box so you can select the multi-polynomial model.

After clicking the red box, select the default model and click on the duck object. This will automatically segment the image. Then, click on the next part and save it. You will then see the left side marked in a red box, where you can see the class name.

Click the save&enter option. annotate the all the images.

Add the images for YOLOv8 format. On the right side, you will see the option to add images in the annotation section. Here, two parts are created: one for annotated images and one for unannotated images.

• First, click the left-side “annotate” option then add the images to the dataset.
• Then click the next “Add Images“.

Now last, we create the dataset so click the “Generate” option on the left side then check the option and press the conitune option.

Then you get the UI of the dataset split option here you can check the train, test and val folders their images automatically split. and click the above red box Export Dataset option and download the zip file. the zip file folder structure is like…

root_file.zip
│
├── test
│   ├── Images
│   └── labels
│
├── train
│   ├── Images
│   └── labels
│
├── val
│   ├── Images
│   └── labels
│
├── data.yaml
└── Readme.roboflow.txt

Step6: Write the script for training the image segmentation model

In this part first, you create the Google Collab file using Drive then upload your dataset. and moute the Google Drive using Google Collab.

1. Use this command for Mount Google Drive

from google.colab import drive

drive.mount('/content/gdrive')

2. Define data directory Use the Constant variable.

DATA_DIR = '/content/drive/MyDrive/YoloV8/Data/'

3. Installing the required package, Install ultralytics

!pip install ultralytics

4. Importing the libraries

import os

from ultralytics import YOLO

5. Load pre-trained YOLOv8 model(here we have different model also check the official documentation there you can see the different model)

model = YOLO('yolov8n-seg.pt') 
 # load a pretrained model (recommended for training)

6. Train the Model

model.train(data="/content/drive/MyDrive/YoloV8/Data/data.yaml", epochs=2, imgsz=640)

# Update the path & and join this line together 

No check your drive The model name folder is created and there model is saved for the prediction we want this model.

7. Predict the Model

#Update the path
model_path="/content/drive/MyDrive/YoloV8/Model/train2/weights/last.pt"
#Update the path
image_path="/content/drive/MyDrive/YoloV8/Data/val/1be566eccffe9561.png"

img = cv2.imread(image_path)
H, W, _ = img.shape

model = YOLO(model_path)

results = model(img)

for result in results:
    for j, mask in enumerate(result.masks.data):

        mask = mask.numpy() * 255

        mask = cv2.resize(mask, (W, H))

        cv2.imwrite('./output.png', mask)

Here you can see the segmentation image is saved.

Now finally we can build both live detection and image segmentation models.

Conclusion

In this blog, we explore live object detection and image segmentation with YOLOv8. For live detection, we import a pre-trained YOLOv8 model and utilize the computer vision library, OpenCV, to open the camera and detect objects. Additionally, we create a Streamlit application for an attractive user interface.

Next, we delve into image segmentation with YOLOv8. We import a pre-trained model and perform transfer learning on a custom dataset. Prior to this, we explored Roboflow for dataset annotation, providing an easy-to-use alternative to tools like LabelImg.

Finally, we predict an image containing a duck. Although the object in the image appears to be a bird, we specify the class name as “duck” for demonstration purposes.

Key Takeaways

• Learning about object detection models like Faster R-CNN, SSD, and the latest YOLOv8.
• Understanding the annotation tool Roboflow and its role in creating datasets for YOLOv8 segmentation models.
• Exploring live object detection using OpenCV (cv2) and Supervision, enhancing practical skills.
• Training and deploying a segmentation model using YOLOv8, gaining hands-on experience.

Frequently Asked Questions