Implementation:Cleanlab Cleanlab OD Visualize

API	`object_detection.summary.visualize`
Source	`cleanlab/object_detection/summary.py:L335-346`
Domains	Machine_Learning, Data_Quality, Object_Detection
Last Updated	2026-02-09

Overview

Implementation of bounding box visualization for object detection label quality review. Renders ground truth labels and model predictions on images with color-coded overlays, supporting both interactive display and file export.

Description

This function renders an annotated image showing ground truth bounding boxes and/or model predictions. It supports:

Multiple image input formats: file path (str), numpy array, or PIL Image.
Optional ground truth label overlay with class names.
Optional prediction overlay filtered by a confidence threshold.
Overlay mode that composites both sets of boxes on a single image.
Configurable figure size and the ability to save to disk.
Class name mapping for human-readable labels on boxes.

The function uses matplotlib for rendering and can display the result inline (e.g., in Jupyter notebooks) or save it to a specified file path.

Usage

This function is used after identifying images with label issues to visually inspect and confirm detected errors. It is typically called in a notebook environment for interactive exploration or in a script to batch-generate annotated images for review.

Code Reference

Source Location

cleanlab/object_detection/summary.py, lines 335-346.

Signature

def visualize(
    image: Union[str, np.ndarray, "Image.Image"],
    *,
    label: Optional[Dict[str, Any]] = None,
    prediction: Optional[np.ndarray] = None,
    prediction_threshold: Optional[float] = None,
    overlay: bool = True,
    class_names: Optional[Dict] = None,
    figsize: Optional[Tuple[int, int]] = None,
    save_path: Optional[str] = None,
    **kwargs,
) -> None

Import

from cleanlab.object_detection.summary import visualize

I/O Contract

Inputs

Parameter	Type	Description
`image`	`Union[str, np.ndarray, Image.Image]`	The image to visualize. Can be a file path string, a numpy array (H,W,C), or a PIL Image object.
`label`	`Optional[Dict[str, Any]]`	Ground truth annotation dictionary with `"bboxes"` (np.ndarray M,4 in xyxy format) and `"labels"` (np.ndarray M,). If None, no ground truth boxes are drawn.
`prediction`	`Optional[np.ndarray]`	Model predictions array of shape (P, K+5). If None, no prediction boxes are drawn.
`prediction_threshold`	`Optional[float]`	Minimum confidence score for displaying predicted boxes. Predictions below this threshold are hidden. If None, all predictions are shown.
`overlay`	`bool`	If True, renders both ground truth and predictions on the same image. If False, renders them side by side. Defaults to True.
`class_names`	`Optional[Dict]`	Mapping from integer class indices to human-readable class names. Used to annotate boxes with class labels.
`figsize`	`Optional[Tuple[int, int]]`	Figure size in inches as (width, height). If None, matplotlib defaults are used.
`save_path`	`Optional[str]`	File path to save the rendered image. If None, the image is only displayed.
`**kwargs`		Additional keyword arguments passed to the underlying rendering functions.

Outputs

Type	Description
`None`	The function does not return a value. It displays a matplotlib figure and/or saves the rendered image to disk.

Usage Examples

import numpy as np
from cleanlab.object_detection.summary import visualize

# Visualize an image with ground truth labels only
label = {
    "bboxes": np.array([[10, 20, 50, 60], [100, 110, 200, 210]]),
    "labels": np.array([0, 1]),
}
visualize("path/to/image.jpg", label=label)

# Visualize with both ground truth and predictions
prediction = np.array([
    [10, 20, 50, 60, 0.9, 0.85, 0.10, 0.05],
    [100, 110, 200, 210, 0.8, 0.05, 0.90, 0.05],
    [150, 150, 250, 250, 0.3, 0.70, 0.20, 0.10],
])

class_names = {0: "cat", 1: "dog", 2: "bird"}

visualize(
    "path/to/image.jpg",
    label=label,
    prediction=prediction,
    prediction_threshold=0.5,
    overlay=True,
    class_names=class_names,
    figsize=(12, 8),
)

# Save the visualization to a file
visualize(
    "path/to/image.jpg",
    label=label,
    prediction=prediction,
    class_names=class_names,
    save_path="output/annotated_image.png",
)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment