SPEK - Simple Python Extraction Kit

SPEK is a simple Python library for real-time object detection using YOLOv8 and OpenCV.
It supports webcam, video, or static image detection and is designed for ease of use.

Features

• Live detection from webcam or video files
• Static image detection
• Filter detection by specific class (e.g., person, dog, car)
• Headless mode (no preview window) for servers
• Callbacks on detection for custom behavior
• Quick-start helper functions

Installation

pip install spek

Usage/Examples

1 Live webcam detection (default)

import spek
# model sizes are n, s, m, l, x
# Detect "person" using YOLOv8s model on webcam
# target_class="person" targets a person
spek.run_live_detection(target_class="person", model_name_or_size="s")

2 Detect a specific target class and run a function

from spek import detect_objects

# Define the callback function
def your_function(info):
    print("Detected person")

# Run detection for "person"
detect_objects(
    target_class="person",
    on_detected=your_function,
    source=0,
    model_size="s"
)

or

from spek import detect_objects

# Define the callback function
def your_function(info):
    print("Detected person")

# Run detection for "person"
detect_objects(target_class="person", on_detected=your_function, source=0, model_size="s")

3 Detect objects from a video file

from spek import detect_objects

# Detect objects from a local video file
detect_objects(source="video.mp4", model_size="m")

4 Detect objects from a static image

from spek import detect_objects

# Detect objects in a single image
detect_objects(source="image.jpg", model_size="l")

5 Run in headless mode (no display, useful for servers)

from spek import detect_objects

# Process webcam feed without displaying window
detect_objects(source=0, model_size="s", headless=True)

6 CLI usage After installation, you can run live detection from the command line:

python -m spek --source 0 --target person --model s

7 Static image via CLI

python -m spek --image input.jpg --target dog --model m

Authors

• @Rudransh Joshi

License

MIT