Improving Labelme Maintainability and Usability for Industrial and Medical Annotation Workflows

[kancheng]

@wkentaro OwO///🐨🐨🐨🐨

A. PR Separation and Maintainability

To improve clarity and maintainability, I have split the previous pull request into two independent and more focused contributions.

The first part is the Traditional Chinese (zh-TW) language support, which has been isolated into its own branch. This feature has been fully tested and is currently stable.

The second part is the drawing-based annotation functionality. This feature has been tested on both Windows and Linux environments and is also generally stable in its current form.

By separating these features into individual PRs, each contribution becomes easier to review, discuss, and iterate on independently. This separation should reduce review overhead and make future maintenance more straightforward.

B. Background and Motivation

My academic background is in medical image segmentation at Peking University, and I am also involved in computer vision applications for industrial production environments. These two domains share similar practical requirements, particularly in annotation workflows used in hospitals and production lines. This motivates me to gradually extend Labelme with features that address real-world usage scenarios rather than purely research-oriented needs.

Traditional Chinese is my native language. I primarily use Mandarin with a Taipei accent and live and work across regions including Hong Kong, Shenzhen, and Macau. Based on this experience, I am familiar with the user habits and localization expectations in these regions, which is why I am particularly interested in improving Traditional Chinese support and overall usability.

Through participation in open-source communities and university auditing courses, I have also been exposed to French, Japanese, and German. Looking forward, I would like to contribute additional language support for French, Japanese, and German, potentially with assistance from international students at Peking University to help review translations and ensure accuracy.

C. Potential Next Steps

One direction I would like to explore is improving data format handling directly within the GUI, allowing users to perform format conversion through simple interactions rather than external scripts. In many production-line and clinical environments, users do not have strong programming or computer science backgrounds, so intuitive and minimal is critical.

This consideration is also a key reason I began working on the drawing annotation functionality. While automated or AI-assisted labeling is valuable, there is still a strong demand for lightweight, manual, and semi-manual annotation tools in real industrial settings.

In parallel, I plan to continue refining and expanding multi-language support, both in terms of coverage and usability consistency across languages.