Imglab - An image annotation application

Repository

https://github.com/NaturalIntelligence/imglab

History

I found my friend struggling to work on dlib's imglab application as he was supposed to setup complete dlib library and requires to install many OS related dependency. As image annotation don't require the knowledge of programming, I created imglab which can be accessed through web.

Introduction

Imglab it web based tool to label images for object. So that they can be used to train dlib or other object detectors for machine learning purpose. You can also download it for offline usage.

ImgLab is platform independent, runs directly from browser, and no prerequisite is needed. It requires very little CPU and memory.

The best thing is that you can use 3rd party libraries for a fast annotation process, which will save both time and effort.

Initially Imglab was designed specially for dlib users. User can create feature points in a shape. They can drag feature points to order them correctly aligned with dlib format. And if you export data file in dlib format, polygon and circle shapes are saved with their boundary box coordinates. However now this application support more formats. You can currently save data in following formats;

• dlib XML
• dlib pts
• Pascal VOC
• COCO

Tenserflow, YOLO, and others formats are in plan.

Other silent features are:

• Drag or resize any annotation shape.
• Select and delete any annotation shape or landmark points.
• Arrange landmark points in specific order my dragging their label up & down, instead of creating them in a particular order.
• Autosave in browser cache. Export to save on disk.
• Hot keys support for easy switch between images, tools, labeling data, or to access other part of the application. Hence it is more convenient and effort saving.
• Set image opacity to highlight annotation shapes and points.
• Tracking lines and mouse coordinates for precise annotation.
• Zoom for more accuracy

Tutorial

Using Imglab is fairly simple. You have to import the images and start annotating them. Feature points can be drawn inside any other shape. When you select a shape, you can provide annotation detail in right side panel. Finally you can save all the data in your desire format.

Here is a video tutorial for more detail;

Technology Stack

HTML, CSS, Javascript, Riotjs, svg.js

Roadmap

• Supporting all major data formats.
• One click annotation for other type of objects as well.
• Intelligence to guess tags and categories
• Automatically annotate similar/common labels in all images
• And much more to save time and effort and make this process easy

Wanna contribute?

This project welcomes contributors. And it also friendly to first timers as well.