feat(visualization): add bounding box overlay for anomaly regions by Midoriya-w · Pull Request #3532 · open-edge-platform/anomalib

Midoriya-w · 2026-04-13T10:17:18Z

📋 Description

Added add_bounding_boxes_to_image() function to item_visualizer.py to draw bounding boxes around detected anomaly regions.

Fixes #(no issue - new feature proposal)

✨ Changes

🚀 New feature (non-breaking change which adds functionality)

What does this PR do?

Draws red bounding boxes around anomalous regions detected in the heatmap
Filters out noise by ignoring contours smaller than 5x5 pixels
Labels each detected region with "DEFECT" text
Fully configurable: threshold, box color, and line thickness are all parameters

Motivation

Anomaly heatmaps show pixel-level intensity but bounding boxes make defect locations immediately interpretable — especially useful for real-time warehouse/industrial inspection demos and production pipelines.

Testing

Tested on MVTecAD bottle dataset with PaDiM model. Bounding boxes correctly highlight defective regions identified by the anomaly heatmap.

Signed-off-by: dinesh <midoriya54378@gmail.com>

Copilot

Pull request overview

This PR introduces a new visualization utility to draw bounding boxes over anomaly regions derived from an anomaly heatmap, improving interpretability of pixel-level anomaly outputs.

Changes:

Added add_bounding_boxes_to_image() to compute a binary anomaly mask via thresholding and draw contour-based bounding boxes.
Added optional parameters for threshold, box color, and line thickness.

Copilot · 2026-04-13T10:57:54Z

+def add_bounding_boxes_to_image(
+    image: np.ndarray,
+    anomaly_map: np.ndarray,
+    threshold: float = 0.5,
+    box_color: tuple = (255, 0, 0),
+    box_thickness: int = 2,
+) -> np.ndarray:


np is used in type annotations and in the function body, but this module does not import NumPy. This will raise NameError when the module is imported (annotations are evaluated eagerly without from __future__ import annotations). Add import numpy as np at module scope (or enable postponed evaluation of annotations) before using np.ndarray/np.uint8.

Copilot · 2026-04-13T10:57:55Z

+    """Draw bounding boxes around detected anomaly regions.
+
+    Args:
+        image: Original input image as numpy array.
+        anomaly_map: Anomaly heatmap from model output.
+        threshold: Score threshold to consider a region anomalous.
+        box_color: BGR color tuple for bounding boxes.
+        box_thickness: Thickness of bounding box lines.
+


The docstring says box_color is in BGR, but anomalib utilities consistently document box colors as RGB (e.g., src/anomalib/utils/post_processing.py:362-405 expects RGB even though it uses OpenCV). Please align this function’s color convention with the rest of the codebase (update docs/type hints and/or convert the image/color ordering accordingly).

Copilot · 2026-04-13T10:57:55Z

+) -> np.ndarray:
+    """Draw bounding boxes around detected anomaly regions.
+
+    Args:
+        image: Original input image as numpy array.
+        anomaly_map: Anomaly heatmap from model output.
+        threshold: Score threshold to consider a region anomalous.
+        box_color: BGR color tuple for bounding boxes.
+        box_thickness: Thickness of bounding box lines.
+
+    Returns:
+        Image with bounding boxes drawn around anomalous regions.
+    """
+    import cv2
+
+    # Normalize anomaly map to 0-1
+    norm_map = (anomaly_map - anomaly_map.min()) / (anomaly_map.max() - anomaly_map.min() + 1e-8)
+
+    # Threshold to get binary mask
+    binary_mask = (norm_map > threshold).astype(np.uint8) * 255


This function min-max normalizes anomaly_map internally, which changes the meaning of threshold (it becomes relative per-image rather than the model’s score scale). Consider either (a) assuming inputs are already normalized (consistent with compute_mask(...): src/anomalib/utils/post_processing.py:308+ / _superimpose_segmentation_mask: src/anomalib/deploy/inferencers/base_inferencer.py:147+), or (b) adding a normalize flag and documenting that threshold applies after normalization.

Suggested change

) -> np.ndarray:

"""Draw bounding boxes around detected anomaly regions.

Args:

image: Original input image as numpy array.

anomaly_map: Anomaly heatmap from model output.

threshold: Score threshold to consider a region anomalous.

box_color: BGR color tuple for bounding boxes.

box_thickness: Thickness of bounding box lines.

Returns:

Image with bounding boxes drawn around anomalous regions.

"""

import cv2

# Normalize anomaly map to 0-1

norm_map = (anomaly_map - anomaly_map.min()) / (anomaly_map.max() - anomaly_map.min() + 1e-8)

# Threshold to get binary mask

binary_mask = (norm_map > threshold).astype(np.uint8) * 255

normalize: bool = True,

) -> np.ndarray:

"""Draw bounding boxes around detected anomaly regions.

Args:

image: Original input image as numpy array.

anomaly_map: Anomaly heatmap from model output. If ``normalize`` is ``False``,

values are assumed to already be on the intended thresholding scale.

threshold: Score threshold to consider a region anomalous. Applied after min-max

normalization when ``normalize=True``; otherwise applied directly to

``anomaly_map`` values.

box_color: BGR color tuple for bounding boxes.

box_thickness: Thickness of bounding box lines.

normalize: Whether to min-max normalize ``anomaly_map`` before thresholding.

Defaults to ``True`` to preserve the existing visualization behavior.

Returns:

Image with bounding boxes drawn around anomalous regions.

"""

import cv2

score_map = anomaly_map

if normalize:

# Normalize anomaly map to 0-1 before applying the threshold.

score_map = (anomaly_map - anomaly_map.min()) / (anomaly_map.max() - anomaly_map.min() + 1e-8)

# Threshold to get binary mask

binary_mask = (score_map > threshold).astype(np.uint8) * 255

Copilot · 2026-04-13T10:57:55Z

+    for contour in contours:
+        x, y, w, h = cv2.boundingRect(contour)
+        if w > 5 and h > 5:  # filter tiny noise boxes
+            cv2.rectangle(result, (x, y), (x + w, y + h), box_color, box_thickness)
+            cv2.putText(result, "DEFECT", (x, y - 5),
+                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, box_color, box_thickness)


There are hard-coded values (w > 5 and h > 5 and the "DEFECT" label) that affect output but aren’t configurable. To avoid magic numbers/strings and make the utility reusable, consider parameters like min_box_size and label (or omit the text overlay by default).

Copilot · 2026-04-13T10:57:55Z

                base_image = add_text_to_image(base_image, title, **text_config)
            output_images.append(base_image)

    return create_image_grid(output_images, nrow=len(output_images)) if output_images else None


PEP8/ruff formatting: add the required blank line(s) between the end of visualize_image_item and this new top-level function definition. Also consider reformatting the multi-line cv2.putText(...) call to match the project’s formatter output.

Suggested change

return create_image_grid(output_images, nrow=len(output_images)) if output_images else None

return create_image_grid(output_images, nrow=len(output_images)) if output_images else None

ashwinvaidya17

Thanks for the contribution. I have added a few comments.

ashwinvaidya17 · 2026-04-13T10:55:56Z

            output_images.append(base_image)

    return create_image_grid(output_images, nrow=len(output_images)) if output_images else None
+def add_bounding_boxes_to_image(


Let's move this to functional.py

ashwinvaidya17 · 2026-04-13T10:56:45Z


    return create_image_grid(output_images, nrow=len(output_images)) if output_images else None
+def add_bounding_boxes_to_image(
+    image: np.ndarray,


To be consistent with other utils such as add_text_to_image can we use Image.Image instead

Suggested change

image: np.ndarray,

image: Image.Image,

ashwinvaidya17 · 2026-04-13T10:57:05Z

+    Returns:
+        Image with bounding boxes drawn around anomalous regions.
+    """
+    import cv2


Try to avoid inline imports if possible.

ashwinvaidya17 · 2026-04-13T10:57:55Z

+            cv2.putText(result, "DEFECT", (x, y - 5),
+                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, box_color, box_thickness)
+
+    return result


Can you run pre-commit hooks

Signed-off-by: dinesh <midoriya54378@gmail.com>

…d to functional Signed-off-by: dinesh <midoriya54378@gmail.com>

Midoriya-w · 2026-04-13T16:06:46Z

Thank you for the review @ashwinvaidya17! I've addressed all the comments:

Moved add bounding boxes to image to functional.py
Updated image type hint to Image.Image
Fixed box color docstring to say RGB instead of BGR
Moved cv2 and numpy imports to top of file
Added min box size and label parameters
Added newline at end of file

Please take another look when you get a chance!

Signed-off-by: dinesh <midoriya54378@gmail.com>

Midoriya-w · 2026-04-29T09:43:58Z

Hi @ashwinvaidya17,

Thanks again for your detailed review earlier I’ve addressed all the points and updated the branch.

I’ve also synced with the latest main branch.
Would really appreciate it if you could take another look when you get a chance.

Happy to make further improvements if needed!

Midoriya-w · 2026-05-18T17:27:12Z

Hey @ashwinvaidya17, just wanted to share that I posted about my Anomalib contribution experience on LinkedIn including the learnings from your code review feedback. Thanks again for taking the time to review!

🔗 https://www.linkedin.com/posts/dinesh-ch-480987307_github-open-edge-platformanomalib-an-activity-7462191416337661952-wabB?utm_source=share&utm_medium=member_desktop&rcm=ACoAAE5G1eABdWqL6YEY33vbAdGViF6AAhk1KwM

Midoriya-w requested review from ashwinvaidya17 and samet-akcay as code owners April 13, 2026 10:17

feat(visualization): add bounding box overlay for anomaly regions

14bb8c8

Signed-off-by: dinesh <midoriya54378@gmail.com>

Midoriya-w force-pushed the feature/bounding-box-visualization branch from 68ea8e9 to 14bb8c8 Compare April 13, 2026 10:20

ashwinvaidya17 requested a review from Copilot April 13, 2026 10:53

Copilot started reviewing on behalf of ashwinvaidya17 April 13, 2026 10:54 View session

Copilot AI reviewed Apr 13, 2026

View reviewed changes

ashwinvaidya17 reviewed Apr 13, 2026

View reviewed changes

Midoriya-w added 2 commits April 13, 2026 21:27

feat(visualization): move bounding box function to functional.py

105c7a6

Signed-off-by: dinesh <midoriya54378@gmail.com>

fix(visualization): remove bounding box fn from item_visualizer, move…

8adf68a

…d to functional Signed-off-by: dinesh <midoriya54378@gmail.com>

Merge branch 'main' into feature/bounding-box-visualization

40214f0

ashwinvaidya17 reviewed Apr 16, 2026

View reviewed changes

Comment thread src/anomalib/visualization/image/functional.py

Midoriya-w added 2 commits April 16, 2026 19:22

test(visualization): add bounding box test

0ba494c

Signed-off-by: dinesh <midoriya54378@gmail.com>

Merge branch 'main' into feature/bounding-box-visualization

0fc20b0

Midoriya-w added 2 commits April 29, 2026 15:14

Merge branch 'main' into feature/bounding-box-visualization

eb787d7

Merge branch 'main' into feature/bounding-box-visualization

f1fd97d

	return create_image_grid(output_images, nrow=len(output_images)) if output_images else None
	return create_image_grid(output_images, nrow=len(output_images)) if output_images else None

Conversation

Midoriya-w commented Apr 13, 2026

📋 Description

✨ Changes

What does this PR do?

Motivation

Testing

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

ashwinvaidya17 left a comment

Choose a reason for hiding this comment

Uh oh!

ashwinvaidya17 Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

ashwinvaidya17 Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

ashwinvaidya17 Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

ashwinvaidya17 Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

Midoriya-w commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Midoriya-w commented Apr 29, 2026

Uh oh!

Midoriya-w commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Midoriya-w commented Apr 13, 2026 •

edited

Loading

Midoriya-w commented May 18, 2026 •

edited

Loading