fix: output segmentation mask

[ktro2828]

Description

This PR tackles the issue #3.

This pull request improves the accuracy and visual quality of 2D semantic segmentation results by refining how segmentation masks are post-processed and visualized. The main changes ensure that segmentation masks are properly scaled and aligned with the original images, and that the visualization overlay is more consistent and visually clear.

Semantic segmentation mask post-processing:

• The segmentation mask is now resized to match the original image size after correcting for any letterboxing and scaling, ensuring pixel-perfect alignment between the mask and the input image. This is done by extracting the valid region from the model output and resizing it back to the original dimensions using nearest neighbor interpolation to preserve class labels.
• The OpenCV image processing header (opencv2/imgproc.hpp) was added to support the resizing operations.

Visualization improvements:

• The visualizer now checks if the input image and segmentation mask sizes match, and resizes the mask if needed to ensure consistent overlay.
• The color mask is converted from BGR to RGB for consistency with the input image, and the overlay blending ratios have been adjusted for better clarity, with the input image weighted more heavily than the mask.

How was this PR tested?

ros2 launch pidnet pidnet.launch.xml input/image:=/nuscenes/CAM_FRONT/image

Screencast.from.09-28-2025.08.36.22.AM.webm

Notes for reviewers

None.

Effects on system behavior

None.

[Copilot]

Pull Request Overview

This PR fixes semantic segmentation mask output to ensure proper alignment with original images and improved visualization quality. The changes address letterbox scaling issues where masks were not correctly resized to match input image dimensions.

• Adds post-processing to resize segmentation masks back to original image dimensions after correcting for letterbox scaling
• Improves visualization by ensuring mask and image size consistency and adjusting color conversion and blending ratios
• Adds bounds checking for class IDs to prevent invalid values

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
mmros/src/detector/semantic_segmenter2d.cpp	Implements proper mask resizing logic to handle letterbox scaling and restore original image dimensions
mmrviz/src/visualizer/segmentation2d_visualizer.cpp	Adds size validation and improves color conversion/blending for better visualization quality

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}