Prediction¶

class Prediction(annotation_json, source=None, model_meta=None, name=None, path=None, url=None, project_id=None, dataset_id=None, image_id=None, video_id=None, frame_index=None, api=None, **kwargs)[source]¶

Bases: object

A single prediction result.

The prediction is primarily represented by annotation_json (Supervisely annotation format). If you provide model_meta (a ProjectMeta or its JSON), you can access the parsed Annotation via annotation.

Parameters:

annotation_json¶: Prediction in Supervisely annotation JSON format.
source : str or int, optional¶: Optional source descriptor (path, id, etc.), used as a hint for loading/visualization.
model_meta=None¶: Output meta of the model, required to construct an Annotation.
name : str, optional¶: Optional name for this prediction (used in visualization filenames).
path : str, optional¶: Local path to the source image/video.
url : str, optional¶: URL to the source image/video.
project_id : int, optional¶: Project id associated with the source.
dataset_id : int, optional¶: Dataset id associated with the source.
image_id : int, optional¶: Image id associated with the source.
video_id : int, optional¶: Video id associated with the source (for frame predictions).
frame_index : int, optional¶: Frame index for video predictions.
api=None¶: API client used to download image/frame by id.

Methods

`from_json`	Create `Prediction` from a JSON dict.
`load_image`	Load the source image (or video frame) into a NumPy array.
`to_json`	Serialize prediction to a JSON-compatible dict.
`visualize`	Render prediction on top of the source image/frame.

Attributes

`annotation`	Get the annotation of the prediction.
`boxes`	Get the bounding boxes of the prediction.
`class_idxs`	Get the class indexes of the prediction.
`classes`	Get the classes of the prediction.
`masks`	Get the masks of the prediction.
`scores`	Get the scores of the prediction.
`track_ids`	Get track IDs for each detection.

classmethod from_json(json_data, **kwargs)[source]¶

Create Prediction from a JSON dict.

This helper accepts both styles of keys coming from different backends: annotation_json or annotation.

Parameters:

json_data : dict¶: Source JSON.
kwargs : dict¶: Additional fields to override/extend JSON (e.g. api, model_meta).

Returns:

Prediction instance.

Return type:

Prediction

Raises:

ValueError – If annotation payload is missing.

load_image()[source]¶

Load the source image (or video frame) into a NumPy array.

The loader uses the first available source in the following order: path → url → image_id (requires api or environment config) → video_id + frame_index → video path/url + frame_index.

Returns:: Image (or frame) as H×W×C NumPy array.
Return type:: numpy.ndarray

to_json()[source]¶

Serialize prediction to a JSON-compatible dict.

Returns:: JSON dict.
Return type:: dict

visualize(save_path=None, save_dir=None, color=None, thickness=None, opacity=0.5, draw_tags=False, fill_rectangles=True)[source]¶

Render prediction on top of the source image/frame.

If save_path (file or directory) or save_dir is provided, the visualization is saved. Otherwise the rendered image is returned.

Parameters:

save_path : str, optional¶: Output file path or directory.
save_dir : str, optional¶: Output directory (alias for directory save_path).
color : List[int], optional¶: Optional RGB color for drawing.
thickness : int, optional¶: Optional line thickness.
opacity : float, optional¶: Bitmap opacity.
draw_tags : bool, optional¶: If True, draws tags on objects.
fill_rectangles : bool, optional¶: If True, fills rectangles.

Returns:

Rendered image.

Return type:

numpy.ndarray

property annotation : supervisely.annotation.annotation.Annotation¶: Get the annotation of the prediction.

property boxes¶: Get the bounding boxes of the prediction.

property class_idxs : numpy.ndarray¶: Get the class indexes of the prediction.

property classes¶: Get the classes of the prediction.

property masks¶: Get the masks of the prediction.

property scores¶: Get the scores of the prediction.

property track_ids¶

Get track IDs for each detection. Returns None for detections without tracking.

Returns:: Track IDs for each detection.
Return type:: np.ndarray