Last Updated on 22/04/2026 by Eran Feit
Boost Your Dataset with yolov8 auto-label segmentation and stop wasting time on manual annotations.
YOLOv8 Video Segmentation for Real Projects This tutorial shows how to turn a regular video into a rich, segmented stream using a pre-trained YOLOv8 segmentation model.yolov8n-seg, run segmentation on each frame, overlay masks in real time, save an output video, and export labeled instance frames to disk.
You can download the code here : https://eranfeit.lemonsqueezy.com/buy/aa058654-26ea-4529-8878-4475db2d593f
or : https://ko-fi.com/s/16c72078f3
Link for the post in Medium users : https://medium.com/@feitgemel/boost-your-dataset-with-yolov8-auto-label-segmentation-eb782002e0f4
Master Computer Vision
Follow my latest tutorials and AI insights on my
Personal Blog .
Beginner Complete CV Bootcamp
Foundation using PyTorch & TensorFlow.
Get Started → Interactive Deep Learning with PyTorch
Hands-on practice in an interactive environment.
Start Learning → Advanced Modern CV: GPT & OpenCV4
Vision GPT and production-ready models.
Go Advanced → Part 1 — Getting YOLOv8 Auto-Label Segmentation Ready for Video Segmentation Here we set up the core ingredients: OpenCV for video handling, Ultralytics YOLOv8 for segmentation, and a clean path structure to read your input video and write processed results.yolov8n-seg.pt model, a lightweight segmentation variant that balances speed and accuracy, making it ideal for real-time or near real-time video workloads on common GPUs — and often even CPU.VideoCapture to read frames and a VideoWriter to export the annotated segmentation video alongside your saved instance frames.yolov8 video segmentation tutorial pattern you can adapt to any folder, filename, or deployment scenario.
You can find the video file (along with the code) here : https://ko-fi.com/s/16c72078f3
Here is a sample image from the video file :
Boost Your Dataset with YOLOv8 Auto-Label Segmentation 9 Detailed Walkthrough & Code (Part 1) ### Import OpenCV for handling video streams, image operations, and display windows. import cv2 ### Import the YOLO class from Ultralytics to load and run the YOLOv8 segmentation model. from ultralytics import YOLO ### Import the Annotator helper and a color utility to draw segmentation masks and labels nicely on each frame. from ultralytics . utils . plotting import Annotator , colors ### Import os to manage folders and file paths when saving segmented instances. import os ### Load the pre-trained YOLOv8 nano segmentation model for fast, real-time-friendly inference on video frames. model = YOLO ( ' yolov8n-seg.pt ' ) ### Extract the class names from the underlying model so we can map class IDs to human-readable labels. names = model . model . names ### Print the list of class names to verify which categories the segmentation model can detect. print ( names ) ### Open the input video file that we want to process with YOLOv8 segmentation. cap = cv2 . VideoCapture ( " Best-Semantic-Segmentation-models/Yolo-V8/Autolabel-Segmentation/test.mp4 " ) ### Create a VideoWriter to save an annotated output video with segmentation overlays in MJPG format at 30 FPS. out = cv2 . VideoWriter ( " Best-Semantic-Segmentation-models/Yolo-V8/Autolabel-Segmentation/out.avi " , cv2 . VideoWriter_fourcc ( * ' MJPG ' ), 30 , ( int ( cap . get ( 3 )), int ( cap . get ( 4 )))) ### Initialize a frame counter so we can generate unique filenames for saved segmented frames. numerator = 0 Short summary for Part 1: yolov8 video segmentation tutorial pipeline that feels predictable and production-minded.
Part 2 — Running YOLOv8 Segmentation on Each Frame In this part, we loop through the video frame by frame and apply YOLOv8 segmentation in real time.model.predict, giving you masks, class IDs, and boxes you can use to overlay colored segmentation regions.Annotator to draw segmented regions and labels; at the same time, we organize per-class folders and save labeled frames for further analysis or dataset bootstrapping.yolov8 video segmentation workflow: you see objects, masks, and per-class exports — all from a compact loop.
Detailed Walkthrough & Code (Part 2) ### Start an infinite loop to read and process frames from the input video one by one. while True : ### Read the next frame from the video stream; ret indicates success, img is the frame. ret , img = cap . read () ### Increment the frame counter to keep track of the current frame index for naming outputs. numerator = numerator + 1 ### If reading failed (end of video or error), log a message and exit the loop cleanly. if not ret : print ( " Break the loop " ) break ### Run YOLOv8 segmentation prediction on the current frame to detect objects and generate masks. results = model . predict ( img ) #print(results[0].masks) # just to see mask detection ### Only continue if YOLOv8 actually returned masks for this frame (i.e., at least one segmented object detected). if results [ 0 ]. masks is not None : ### Extract the class IDs for all detected objects and move them to CPU as a Python list. clss = results [ 0 ]. boxes . cls . cpu (). tolist () ### Get the polygon coordinates (xy format) for all segmentation masks in the current frame. masks = results [ 0 ]. masks . xy ### Initialize an Annotator object on the current frame so we can draw masks, labels, and boundaries. annotator = Annotator ( img , line_width = 2 ) ### Iterate over each detected mask and its corresponding class ID together. for idx , ( mask , cls ) in enumerate ( zip ( masks , clss )): ### Convert the numeric class ID into a readable label using the model's class names. det_label = names [ int ( cls )] ### Draw the segmentation mask and a labeled bounding region for this detection using a consistent color. annotator . seg_bbox ( mask = mask , mask_color = colors ( int ( cls ), True ), det_label = det_label ) ### Build a folder path for this object class to organize exported segmented frames per category. instance_folder = os . path . join ( " Best-Semantic-Segmentation-models/Yolo-V8/Autolabel-Segmentation " , det_label ) ### Create the class-specific folder if it does not already exist. os . makedirs ( instance_folder , exist_ok =True ) ### Build a unique filename for this frame's annotated image using the class label and frame index. instance_path = os . path . join ( instance_folder , f " { det_label } _ { str ( numerator ) } .png" ) ### Save the current annotated frame into the class-specific folder for later review or dataset creation. cv2 . imwrite ( instance_path , img ) Notes (kept practical & honest):
As written, each detection saves the full annotated frame under that class folder (not a cropped-only instance). This loop is the heart of the yolov8 instance segmentation video pipeline: detect, segment, visualize, store. YOLOv8 Auto-Label Segmentation – the main core for segmentation Short summary for Part 2:
Part 3 — Displaying, Saving, and Cleaning Up Gracefully The last part keeps things user-friendly and production-safe.
Detailed Walkthrough & Code (Part 3) ### Show the current annotated frame in a live window so you can monitor segmentation results in real time. cv2 . imshow ( " img " , img ) ### Write the annotated frame into the output video file for later playback or sharing. out . write ( img ) ### Listen for a key press; if the user presses 'q', stop processing early and break the loop. if cv2 . waitKey ( 1 ) & 0x FF == ord ( ' q ' ): break ### Release the video capture resource once processing is complete. cap . release () ### Release the video writer resource to finalize and close the output video file. out . release () ### Close all OpenCV display windows to clean up the GUI and free system resources. cv2 . destroyAllWindows () Short summary for Part 3: python yolov8 segmentation example that behaves like real production code, not just a quick script.
Here is the result :
Boost Your Dataset with YOLOv8 Auto-Label Segmentation 10 FAQ : What is YOLOv8 video segmentation in this tutorial? It applies a YOLOv8 segmentation model to each video frame to draw masks, labels, and save annotated outputs for real-time analysis or dataset creation.
Do I need my own dataset to start? No, you can begin with the pre-trained yolov8n-seg model and later fine-tune on a custom dataset if your objects are not covered by COCO classes.
Why are frames saved in class-specific folders? Saving by class keeps results organized, making it easy to review predictions and reuse frames for training or documentation.
Can this pipeline run in real time? With yolov8n-seg and reasonable resolutions, many setups can approach real-time, especially when using a GPU for inference.
How do I stop the script while it runs? Press the ‘q’ key in the OpenCV window to break the loop and close the video stream gracefully.
Can I use a webcam instead of a video file? Yes, replace the video path in VideoCapture with your webcam index and keep the rest of the segmentation loop unchanged.
How can I save cropped segmented objects only? Use each mask or bounding box to crop the region from the frame before saving, instead of writing the full annotated image.
Is MJPG the only supported codec? No, it is a safe default; you can switch to any codec supported by your system if you prefer smaller or different output formats.
Can I filter detections by specific classes? Yes, simply keep only the detections whose class IDs match the categories you care about before drawing or saving.
Is this YOLOv8 video segmentation setup production-ready? It is a strong baseline; for production, extend it with configuration files, logging, monitoring, and fine-tuned models if needed.
Conclusion YOLOv8 video segmentation gives you much more than pretty overlays. You save time using YOLOv8 for Auto-Label Segmentationyolov8n-seg model, a simple loop, and intuitive folder naming so anyone comfortable with basic Python can follow along.
From here, you can tune every step without rewriting the core idea.
Most importantly, this yolov8 video segmentation tutorial shows a mindset: start simple, keep things readable, and let the code serve real use cases like surveillance analytics, sports breakdowns, traffic monitoring, or automatic content tagging.
Connect : ☕ Buy me a coffee — https://ko-fi.com/eranfeit
🖥️ Email : feitgemel@gmail.com
🌐 https://eranfeit.net
🤝 Fiverr : https://www.fiverr.com/s/mB3Pbb
Enjoy,
Eran
Skip to content 
