Table of Contents

• Installation
• Usage
• Test
• License

Installation

pip install bluevision

Usage

Simple video demo

from bluevision.demo import video_demo

video_demo(
    weights="yolov8s-coco.safetensors",
    video_path="sample.mp4",
    model_size="s",
    track=True,
    show=True,
    save_path="output.mp4",
)

Video Inference Process

import cv2
import supervision as sv
import bluevision as bv
from bluevision.utils import to_supervision_detections, make_labels

# Initialize
detector = bv.solutions.Detector(model=bv.solutions.detector.models.Yolov8(size='s'),
                                 nms=bv.utils.nms.soft_nms,
                                 weights="yolov8s.safetensors")
tracker = bv.utils.tracker.BYTETracker(track_thresh=0.15, match_thresh=0.9,
                                       track_buffer=60, frame_rate=30)
box_annotator = sv.BoundingBoxAnnotator(thickness=2)
label_annotator = sv.LabelAnnotator(text_scale=0.5, text_padding=2)

# Load sample video
vid = cv2.VideoCapture('sample.mp4')

# Start
while True:
    ret, original_image = vid.read()
    if not ret:
        break

    detections = detector(original_image)
    detections = tracker.update(detections)

    # Draw bboxes using supervision
    sv_detections = to_supervision_detections(detections)
    annotated_frame = box_annotator.annotate(
        scene=original_image,
        detections=sv_detections,
    )
    annotated_frame = label_annotator.annotate(
        scene=annotated_frame,
        detections=sv_detections,
        labels=make_labels(sv_detections)
    )

    cv2.imshow('annotated image', annotated_frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
cv2.destroyAllWindows()

BlueVision

Solution

Object Detector

Utils

NMS Object Tracker

Test

$ python test_with_time.py
Using device: mps
preprocess: 0.00225s, infer: 0.015508s, postprocess: 0.01313s, track: 0.00108s, draw: 0.000670s, total: 0.03462s, t-avg: 0.03576s
total frame : 1050
total elapsed time: 56.25610s
total inference time: 37.55273s

License

bluevision is distributed under the terms of the MIT license.