supervision-0.14.0

🚀 Added

• Support for SAHI inference technique with sv.InferenceSlicer. (#282)

>>> import cv2
>>> import supervision as sv
>>> import numpy as np
>>> from ultralytics import YOLO

>>> image = cv2.imread(SOURCE_IMAGE_PATH)
>>> model = YOLO(...)

>>> def callback(image_slice: np.ndarray) -> sv.Detections:
...     result = model(image_slice)[0]
...     return sv.Detections.from_ultralytics(result)

>>> slicer = sv.InferenceSlicer(callback = callback)

>>> detections = slicer(image)

inference-slicer.mov

• Detections.from_deepsparse to enable seamless integration with DeepSparse framework. (#297)

• sv.Classifications.from_ultralytics to enable seamless integration with Ultralytics framework. This will enable you to use supervision with all models that Ultralytics supports. (#281)

  Warning
  sv.Detections.from_yolov8 and sv.Classifications.from_yolov8 are now deprecated and will be removed with supervision-0.16.0 release.

• First supervision usage example script showing how to detect and track objects on video using YOLOv8 + Supervision. (#341)

detect-and-track-objects-on-video.mov

🌱 Changed

• sv.ClassificationDataset and sv.DetectionDataset now use image path (not image name) as dataset keys. (#296)

🛠️ Fixed

• Detections.from_roboflow to filter out polygons with less than 3 points. (#300)

🏆 Contributors

@hardikdava (Hardik Dava), @onuralpszr (Onuralp SEZER), @mayankagarwals (Mayank Agarwal), @rizavelioglu (Riza Velioglu), @arjun-234 (Arjun D.), @mwitiderrick (Derrick Mwiti), @ShubhamKanitkar32, @gasparitiago (Tiago De Gaspari), @capjamesg (James Gallagher), @SkalskiP (Piotr Skalski)

supervision-0.13.0

🚀 Added

• Support for mean average precision (mAP) for object detection models with sv.MeanAveragePrecision. (#236)

>>> import supervision as sv
>>> from ultralytics import YOLO

>>> dataset = sv.DetectionDataset.from_yolo(...)

>>> model = YOLO(...)
>>> def callback(image: np.ndarray) -> sv.Detections:
...     result = model(image)[0]
...     return sv.Detections.from_yolov8(result)

>>> mean_average_precision = sv.MeanAveragePrecision.benchmark(
...     dataset = dataset,
...     callback = callback
... )

>>> mean_average_precision.map50_95
0.433

• Support for ByteTrack for object tracking with sv.ByteTrack. (#256)

>>> import supervision as sv
>>> from ultralytics import YOLO

>>> model = YOLO(...)
>>> byte_tracker = sv.ByteTrack()
>>> annotator = sv.BoxAnnotator()

>>> def callback(frame: np.ndarray, index: int) -> np.ndarray:
...     results = model(frame)[0]
...     detections = sv.Detections.from_yolov8(results)
...     detections = byte_tracker.update_from_detections(detections=detections)
...     labels = [
...         f"#{tracker_id} {model.model.names[class_id]} {confidence:0.2f}"
...         for _, _, confidence, class_id, tracker_id
...         in detections
...     ]
...     return annotator.annotate(scene=frame.copy(), detections=detections, labels=labels)

>>> sv.process_video(
...     source_path='...',
...     target_path='...',
...     callback=callback
... )

byte_track_result_small.mp4

• sv.Detections.from_ultralytics to enable seamless integration with Ultralytics framework. This will enable you to use supervision with all models that Ultralytics supports. (#222)

  Warning
  sv.Detections.from_yolov8 is now deprecated and will be removed with supervision-0.15.0 release.

• sv.Detections.from_paddledet to enable seamless integration with PaddleDetection framework. (#191)

• Support for loading PASCAL VOC segmentation datasets with sv.DetectionDataset.. (#245)

🏆 Contributors

@hardikdava (Hardik Dava), @kirilllzaitsev (Kirill Zaitsev), @onuralpszr (Onuralp SEZER), @dbroboflow, @mayankagarwals (Mayank Agarwal), @danigarciaoca (Daniel M. García-Ocaña), @capjamesg (James Gallagher), @SkalskiP (Piotr Skalski)

supervision-0.12.0

Warning
With the supervision-0.12.0 release, we are terminating official support for Python 3.7. (#179)

🚀 Added

• Initial support for object detection model benchmarking with sv.ConfusionMatrix. (#177)

>>> import supervision as sv
>>> from ultralytics import YOLO

>>> dataset = sv.DetectionDataset.from_yolo(...)

>>> model = YOLO(...)
>>> def callback(image: np.ndarray) -> sv.Detections:
...     result = model(image)[0]
...     return sv.Detections.from_yolov8(result)

>>> confusion_matrix = sv.ConfusionMatrix.benchmark(
...     dataset = dataset,
...     callback = callback
... )

>>> confusion_matrix.matrix
array([
    [0., 0., 0., 0.],
    [0., 1., 0., 1.],
    [0., 1., 1., 0.],
    [1., 1., 0., 0.]
])

• Detections.from_mmdetection to enable seamless integration with MMDetection framework. (#173)

• Ability to install package in headless or desktop mode. (#130)

🌱 Changed

• Packing method from setup.py to pyproject.toml. (#180)

🛠️ Fixed

• sv.DetectionDataset.from_cooc can't be loaded when there are images without annotations. (#188)
• sv.DetectionDataset.from_yolo can't load background instances. (#226)

🏆 Contributors

@kirilllzaitsev @hardikdava @onuralpszr @Ucag @SkalskiP @capjamesg

supervision-0.11.1

🛠️ Fixed

• as_folder_structure fails to save sv.ClassificationDataset when it is result of inference. (#165)

🏆 Contributors

@capjamesg @SkalskiP

supervision-0.11.0

🚀 Added

• Ability to load and save sv.DetectionDataset in COCO format using as_coco and from_coco methods. (#150)

>>> import supervision as sv

>>> ds = sv.DetectionDataset.from_coco(
...     images_directory_path='...',
...     annotations_path='...'
... )

>>> ds.as_coco(
...     images_directory_path='...',
...     annotations_path='...'
... )

• Ability to marge multiple sv.DetectionDataset together using merge method. (#158)

>>> import supervision as sv

>>> ds_1 = sv.DetectionDataset(...)
>>> len(ds_1)
100
>>> ds_1.classes
['dog', 'person']

>>> ds_2 = sv.DetectionDataset(...)
>>> len(ds_2)
200
>>> ds_2.classes
['cat']

>>> ds_merged = sv.DetectionDataset.merge([ds_1, ds_2])
>>> len(ds_merged)
300
>>> ds_merged.classes
['cat', 'dog', 'person']

• Additional start and end arguments to sv.get_video_frames_generator allowing to generate frames only for a selected part of the video. (#162)

🛠️ Fixed

• Incorrect loading of YOLO dataset class names from data.yaml. (#157)

🏆 Contributors

@SkalskiP @hardikdava

supervision-0.10.0

🚀 Added

• Ability to load and save sv.ClassificationDataset in a folder structure format. (#125)

>>> import supervision as sv

>>> cs = sv.ClassificationDataset.from_folder_structure(
...     root_directory_path='...'
... )

>>> cs.as_folder_structure(
...     root_directory_path='...'
... )

• Support for sv.ClassificationDataset.split allowing to divide sv.ClassificationDataset into two parts. (#125)

>>> import supervision as sv

>>> cs = sv.ClassificationDataset(...)
>>> train_cs, test_cs = cs.split(split_ratio=0.7, random_state=42, shuffle=True)

>>> len(train_cs), len(test_cs)
(700, 300)

• Ability to extract masks from Roboflow API results using sv.Detections.from_roboflow. (#110)

• Supervision Quickstart notebook where you can learn more about Detection, Dataset and Video APIs.

🌱 Changed

• sv.get_video_frames_generator documentation to better describe actual behavior. (#135)

🏆 Contributors

@capjamesg @dankresio @SkalskiP

supervision-0.9.0

🚀 Added

• Ability to select sv.Detections by index, list of indexes or slice. Here is an example illustrating the new selection methods. (#118)

>>> import supervision as sv

>>> detections = sv.Detections(...)
>>> len(detections[0])
1
>>> len(detections[[0, 1]])
2
>>> len(detections[0:2])
2

• Ability to extract masks from YOLOv8 results using sv.Detections.from_yolov8. Here is an example illustrating how to extract boolean masks from the result of the YOLOv8 model inference. (#101)

>>> import cv2
>>> from ultralytics import YOLO
>>> import supervision as sv

>>> image = cv2.imread(...)
>>> image.shape
(640, 640, 3)

>>> model = YOLO('yolov8s-seg.pt')
>>> result = model(image)[0]
>>> detections = sv.Detections.from_yolov8(result)
>>> detections.mask.shape
(2, 640, 640)

• Ability to crop the image using sv.crop. Here is an example showing how to get a separate crop for each detection in sv.Detections. (#122)

>>> import cv2
>>> import supervision as sv

>>> image = cv2.imread(...)
>>> detections = sv.Detections(...)
>>> len(detections)
2
>>> crops = [
...     sv.crop(image=image, xyxy=xyxy) 
...     for xyxy 
...     in detections.xyxy
... ]
>>> len(crops)
2

• Ability to conveniently save multiple images into directory using sv.ImageSink. An example shows how to save every tenth video frame as a separate image. (#120)

>>> import supervision as sv

>>> with sv.ImageSink(target_dir_path='target/directory/path') as sink:
...     for image in sv.get_video_frames_generator(source_path='source_video.mp4', stride=10):
...         sink.save_image(image=image)

🛠️ Fixed

• Inconvenient handling of sv.PolygonZone coordinates. Now sv.PolygonZone accepts coordinates in the form of [[x1, y1], [x2, y2], ...] that can be both integers and floats. (#106)

🏆 Contributors

@SkalskiP @lomnes-atlast-food @hardikdava

supervision-0.8.0

🚀 Added

• Support for dataset inheritance. The current Dataset got renamed to DetectionDataset. Now DetectionDataset inherits from BaseDataset. This change was made to enforce the future consistency of APIs of different types of computer vision datasets. (#100)
• Ability to save datasets in YOLO format using DetectionDataset.as_yolo. (#100)

>>> import supervision as sv

>>> ds = sv.DetectionDataset(...)
>>> ds.as_yolo(
...     images_directory_path='...',
...     annotations_directory_path='...',
...     data_yaml_path='...'
... )

• Support for DetectionDataset.split allowing to divide DetectionDataset into two parts. (#102)

>>> import supervision as sv

>>> ds = sv.DetectionDataset(...)
>>> train_ds, test_ds = ds.split(split_ratio=0.7, random_state=42, shuffle=True)

>>> len(train_ds), len(test_ds)
(700, 300)

🌱 Changed

• Default value of approximation_percentage parameter from 0.75 to 0.0 in DetectionDataset.as_yolo and DetectionDataset.as_pascal_voc. (#100)

🏆 Contributors

• @SkalskiP

supervision-0.7.0

🚀 Added

• Detections.from_yolo_nas to enable seamless integration with YOLO-NAS model. (#91)
• Ability to load datasets in YOLO format using Dataset.from_yolo. (#86)
• Detections.merge to merge multiple Detections objects together. (#84)

🌱 Changed

• LineZoneAnnotator.annotate to allow for the custom text for the in and out tags. (#44)

🛠️ Fixed

• LineZoneAnnotator.annotate does not return annotated frame. (#81)

🏆 Contributors

• @SkalskiP
• @iPoe
• @hardikdava

supervision-0.6.0

🚀 Added

• Initial Dataset support and ability to save Detections in Pascal VOC XML format. (#71)
• New mask_to_polygons, filter_polygons_by_area, polygon_to_xyxy and approximate_polygon utilities. (#71)
• Ability to load Pascal VOC XML object detections dataset as Dataset. (#72)

🌱 Changed

• order of Detections attributes to make it consistent with order of objects in __iter__ tuple. (#70)
• generate_2d_mask to polygon_to_mask. (#71)

🏆 Contributors

• @SkalskiP
• @alexandercarruthers