Skip to content


YOLOv8 predict mode can generate predictions for various tasks, returning either a list of Results objects or a memory-efficient generator of Results objects when using the streaming mode. Enable streaming mode by passing stream=True in the predictor's call method.


inputs = [img, img]  # list of numpy arrays
results = model(inputs)  # list of Results objects

for result in results:
    boxes = result.boxes  # Boxes object for bbox outputs
    masks = result.masks  # Masks object for segmentation masks outputs
    probs = result.probs  # Class probabilities for classification outputs
inputs = [img, img]  # list of numpy arrays
results = model(inputs, stream=True)  # generator of Results objects

for result in results:
    boxes = result.boxes  # Boxes object for bbox outputs
    masks = result.masks  # Masks object for segmentation masks outputs
    probs = result.probs  # Class probabilities for classification outputs


Streaming mode with stream=True should be used for long videos or large predict sources, otherwise results will accumuate in memory and will eventually cause out-of-memory errors.


YOLOv8 can accept various input sources, as shown in the table below. This includes images, URLs, PIL images, OpenCV, numpy arrays, torch tensors, CSV files, videos, directories, globs, YouTube videos, and streams. The table indicates whether each source can be used in streaming mode with stream=True ✅ and an example argument for each source.

source model(arg) type notes
image 'im.jpg' str, Path
URL '' str
screenshot 'screen' str
PIL'im.jpg') PIL.Image HWC, RGB
OpenCV cv2.imread('im.jpg')[:,:,::-1] np.ndarray HWC, BGR to RGB
numpy np.zeros((640,1280,3)) np.ndarray HWC
torch torch.zeros(16,3,320,640) torch.Tensor BCHW, RGB
CSV 'sources.csv' str, Path RTSP, RTMP, HTTP
video ✅ 'vid.mp4' str, Path
directory ✅ 'path/' str, Path
glob ✅ 'path/*.jpg' str Use * operator
YouTube ✅ '' str
stream ✅ 'rtsp://' str RTSP, RTMP, HTTP

Image and Video Formats

YOLOv8 supports various image and video formats, as specified in yolo/data/ See the tables below for the valid suffixes and example predict commands.

Image Suffixes

Image Suffixes Example Predict Command Reference
.bmp yolo predict source=image.bmp Microsoft BMP File Format
.dng yolo predict source=image.dng Adobe DNG
.jpeg yolo predict source=image.jpeg JPEG
.jpg yolo predict source=image.jpg JPEG
.mpo yolo predict source=image.mpo Multi Picture Object
.png yolo predict source=image.png Portable Network Graphics
.tif yolo predict source=image.tif Tag Image File Format
.tiff yolo predict source=image.tiff Tag Image File Format
.webp yolo predict source=image.webp WebP
.pfm yolo predict source=image.pfm Portable FloatMap

Video Suffixes

Video Suffixes Example Predict Command Reference
.asf yolo predict source=video.asf Advanced Systems Format
.avi yolo predict source=video.avi Audio Video Interleave
.gif yolo predict source=video.gif Graphics Interchange Format
.m4v yolo predict source=video.m4v MPEG-4 Part 14
.mkv yolo predict source=video.mkv Matroska
.mov yolo predict QuickTime File Format
.mp4 yolo predict source=video.mp4 MPEG-4 Part 14 - Wikipedia
.mpeg yolo predict source=video.mpeg MPEG-1 Part 2
.mpg yolo predict source=video.mpg MPEG-1 Part 2
.ts yolo predict source=video.ts MPEG Transport Stream
.wmv yolo predict source=video.wmv Windows Media Video
.webm yolo predict source=video.webm WebM Project

Working with Results

The Results object contains the following components:

  • Results.boxes: Boxes object with properties and methods for manipulating bounding boxes
  • Results.masks: Masks object for indexing masks or getting segment coordinates
  • Results.probs: torch.Tensor containing class probabilities or logits
  • Results.orig_img: Original image loaded in memory
  • Results.path: Path containing the path to the input image

Each result is composed of a torch.Tensor by default, which allows for easy manipulation:


results = results.cuda()
results = results.cpu()
results ='cpu')
results = results.numpy()


Boxes object can be used to index, manipulate, and convert bounding boxes to different formats. Box format conversion operations are cached, meaning they're only calculated once per object, and those values are reused for future calls.

  • Indexing a Boxes object returns a Boxes object:


results = model(img)
boxes = results[0].boxes
box = boxes[0]  # returns one box
  • Properties and conversions

Boxes Properties

boxes.xyxy  # box with xyxy format, (N, 4)
boxes.xywh  # box with xywh format, (N, 4)
boxes.xyxyn  # box with xyxy format but normalized, (N, 4)
boxes.xywhn  # box with xywh format but normalized, (N, 4)
boxes.conf  # confidence score, (N, 1)
boxes.cls  # cls, (N, 1)  # raw bboxes tensor, (N, 6) or boxes.boxes


Masks object can be used index, manipulate and convert masks to segments. The segment conversion operation is cached.


results = model(inputs)
masks = results[0].masks  # Masks object
masks.xy  # x, y segments (pixels), List[segment] * N
masks.xyn  # x, y segments (normalized), List[segment] * N  # raw masks tensor, (N, H, W) or masks.masks 


probs attribute of Results class is a Tensor containing class probabilities of a classification operation.


results = model(inputs)
results[0].probs  # cls prob, (num_class, )

Class reference documentation for Results module and its components can be found here

Plotting results

You can use plot() function of Result object to plot results on in image object. It plots all components(boxes, masks, classification logits, etc.) found in the results object


res = model(img)
res_plotted = res[0].plot()
cv2.imshow("result", res_plotted)
  • show_conf (bool): Show confidence
  • line_width (Float): The line width of boxes. Automatically scaled to img size if not provided
  • font_size (Float): The font size of . Automatically scaled to img size if not provided

Streaming Source for-loop

Here's a Python script using OpenCV (cv2) and YOLOv8 to run inference on video frames. This script assumes you have already installed the necessary packages (opencv-python and ultralytics).

Streaming for-loop

import cv2
from ultralytics import YOLO

# Load the YOLOv8 model
model = YOLO('')

# Open the video file
video_path = "path/to/your/video/file.mp4"
cap = cv2.VideoCapture(video_path)

# Loop through the video frames
while cap.isOpened():
    # Read a frame from the video
    success, frame =

    if success:
        # Run YOLOv8 inference on the frame
        results = model(frame)

        # Visualize the results on the frame
        annotated_frame = results[0].plot()

        # Display the annotated frame
        cv2.imshow("YOLOv8 Inference", annotated_frame)

        # Break the loop if 'q' is pressed
        if cv2.waitKey(1) & 0xFF == ord("q"):
        # Break the loop if the end of the video is reached

# Release the video capture object and close the display window