supervision

Introduction to the Supervision Project

The Supervision project is designed to streamline and enhance computer vision tasks through a suite of reusable tools. Whether the task involves loading datasets, visualizing detections on images or videos, or counting objects in specified areas, Supervision provides a robust solution tailored to meet these needs.

Installation

To get started with Supervision, users can easily install it via Python's package manager, pip. It requires a Python environment of version 3.8 or higher. Users looking for more detailed installation instructions, including using conda and mamba or installing from source, can refer to the comprehensive online guide.

pip install supervision

Model Integration

Supervision is model-agnostic, allowing it to work seamlessly with any classification, detection, or segmentation model. There are convenient connectors available for popular libraries such as Ultralytics, Transformers, and MMDetection. This flexibility ensures that users can integrate Supervision with their existing models without hassle.

Here is a simple example using the YOLO model from Ultralytics:

import cv2
import supervision as sv
from ultralytics import YOLO

image = cv2.imread(...)
model = YOLO("yolov8s.pt")
result = model(image)[0]
detections = sv.Detections.from_ultralytics(result)

len(detections)  # Output: 5

Annotators

The project provides a variety of highly customizable annotators, enabling users to visualize data effectively based on their unique requirements. An example of using the BoxAnnotator to display detections on an image is shown below:

import cv2
import supervision as sv

image = cv2.imread(...)
detections = sv.Detections(...)

box_annotator = sv.BoxAnnotator()
annotated_frame = box_annotator.annotate(
  scene=image.copy(),
  detections=detections)

Dataset Handling

Supervision includes utilities that simplify the handling of datasets. Users can load, split, merge, and save datasets in various supported formats such as COCO, YOLO, and Pascal VOC. This flexibility makes managing datasets efficient and straightforward.

Below is an example of how to load a dataset in COCO format:

import supervision as sv

ds = sv.DetectionDataset.from_coco(
    images_directory_path="path/to/images",
    annotations_path="path/to/annotations.coco.json",
)

for path, image, annotation in ds:
    # Image is loaded on demand

Learning and Contribution

Supervision offers a wealth of learning resources, including how-to guides, end-to-end examples, cheat sheets, and cookbooks. Whether users are new to computer vision or seasoned practitioners, these resources can help them get the most out of the tools provided by Supervision.

For those interested in contributing to the project, there's a detailed contributing guide available. Contributions are welcomed and celebrated, as they help to expand and enhance the capabilities of Supervision.

Built with Supervision

Many projects have been built using the Supervision tools, and the community is encouraged to share their projects. This not only showcases the versatility of Supervision but also helps inspire others in the computer vision field.

Documentation and Support

Comprehensive documentation is available, providing detailed guidance on how Supervision can accelerate and improve the reliability of computer vision projects. For additional support, users can engage with the community through forums and discussions, making it easier to solve problems and learn from others' experiences.

In summary, the Supervision project offers a powerful suite of computer vision tools that are essential for anyone looking to develop flexible, reliable, and efficient vision-based applications.