Add Magic Wand tool for region selection based on color/pixel similarity (like Photoshop)

[zhangbest5]

Search before asking

• I have searched the X-AnyLabeling issues and found no similar feature requests.

Description

问题描述

目前在做像素级标注时，对于颜色/纹理相近的区域（如天空、草地、墙面等），需要手动点击多边形描点，非常费时。希望增加类似Photoshop的"魔棒"工具。

期望功能

1. 点击图像上某个像素，自动选取颜色相近的连通区域
2. 提供"容差(Tolerance)"滑动条，控制选取敏感度
3. 支持按住Shift追加选区、Alt减去选区
4. 选区可直接转为多边形标注

现有替代方案

目前X-AnyLabeling已集成SAM系列模型，可通过点击正/负点生成分割掩码，但：

• SAM需要加载大模型，启动慢，资源占用高
• SAM基于语义理解，对于纯色连片区域反而不如魔棒直接

建议实现方式

参考OpenCV的 floodFill 算法实现，轻量快速，无需加载AI模型。

感谢开发者！

Problem

When doing pixel-level annotation, selecting large continuous areas (sky, grass, walls, etc.) requires manually clicking many polygon points, which is very time-consuming.

Requested Feature

1. Click on a pixel → automatically select connected region with similar color
2. Add a "Tolerance" slider to control selection sensitivity
3. Support Shift+click to add, Alt+click to subtract from selection
4. Convert the selection directly to a polygon annotation

Current Workaround

X-AnyLabeling already integrates SAM models, which can generate masks via point prompts. However:

• SAM requires loading large models (slow startup, high resource usage)
• For large uniform-color areas, SAM is overkill

Suggested Implementation

A lightweight flood fill algorithm (like OpenCV's floodFill) without requiring AI model loading.

Thanks for considering!

Use case

No response

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[CVHub520]

Thanks for the suggestion. From our side, adapting a brand-new algorithm just for this use case would have pretty low ROI. If compute resources are limited, we’d recommend lightweight interactive segmentation models like MobileSAM or EfficientSAM.

For SAM-based workflows, the encoder is usually preloaded first, and that is the most time-consuming part. After that, it stays warm in memory, so the decoder step is very fast and can usually respond in milliseconds.