GoFireStar/picA

PicA is an AI-powered and user-friendly image annotation tool designed to simplify the process of preparing image datasets for machine learning projects. With a focus on versatility and efficiency, picA offers a range of annotation features tailored to different use cases, including object counting, object detection, instance segmentation, semantic segmentation, and panoptic segmentation.

🌟 Key Features 🌟

Versatile Annotation

picA supports a variety of annotation tasks, allowing you to seamlessly annotate images for different purposes, from simple object counting to complex instance and semantic segmentation.
Deep Learning Integration

Harness the capabilities of deep learning models to enhance your annotation process. picA supports integration with pre-trained models, enabling you to accelerate and improve annotation accuracy.
Custom Model Deployment [TBD]

Utilize your own custom deep learning models within picA. Tailor your annotations to your project's unique requirements, ensuring precise and reliable results.
Export in COCO Format

Easily export your annotations in the COCO (Common Objects in Context) format, a widely used standard in the computer vision community. Your annotated data is ready for integration into your machine learning pipeline.

Roadmap 👀

Wait to Test

COCO format annotations export.
Category visibility control.
Customized model integration.

Short-Term Goals

Nested category supports.
Keypoint annotation supports.
Rotated bounding box annotation supports.

Installlation 😱

See below for a quickstart installation or it's recommended to directly pulling a docker file.

Code Install

Pip installation need Python>=3.8 and requirements below:

Python >= 3.8
pip >= 22.0
PyTorch >= 1.10

cd /path/to/picA
pip install -r requirements.txt

Docker Install

TBD.

Getting Started 🏃

Launch

picA may be launched directly in the Command Line Interface (CLI):

cd /path/to/picA
python main.py

Create or Import a project

To create a new project, click File -> New Project to select a folder that contains image data. PicA officially supports jpg, jpeg, png, and bmp image formats.

Project Name
|--- image01.jpg
|--- image02.jpg
|--- ...

The project will be saved in the selected foler and the results look like:

Project Name
|--- image01.jpg
|--- image02.jpg
|--- ...
|--- annotations
|.. |--- annotations.json
|.. |--- masks
|.. |--- color_masks
|.. |.. |--- image01.png
|.. |.. |--- image02.png
|.. |.. |--- ...

When importing an existing project with File -> Import Project, ensure that you select the Project Name folder and verify that all images and annotations are located within the same Project Name directory.

If creating or importing a project encounters issues, a simple solution is to relaunch the application.

Merge two projects

Annotation integration is also available. Upon opening the Project 1 window, you can opt to Import Project to merge Project 2 with the existing one. However, please exercise caution and consider the following guidelines:

Merging is supported only for projects with the same annotation tasks.
Ensure that all images are situated within the directory of the Project 1.
In cases of conflicts, such as annotations on the same image in both projects, picA will prioritize retaining annotations from Project 1.

All annotations will be preserved within the context of the Project 1, providing a consolidated and organized annotation repository.

Operation

In picA, interactions can be categorized into two distinct operations: File or Category Selection and Annotation Manipulation.

File or Category Selection

This operation is accessible through the side panel. Here, users can seamlessly navigate between different files and efficiently manage categories. Within this panel, you have the capability to create, delete, and modify categories. Additionally, the ability to create instances offers for Instance and Panoptic Segmentation.

Annotation Manipulation

Found at the bottom panel, this operation encompasses the core of annotation activities. It's divided into two primary modes: Select and Draw. The Select mode empowers users to highlight, modify, and delete existing annotations, as well as reassign them to different categories. On the other hand, the Draw mode employs distinct terminology tailored to specific annotation tasks, such as Click, Rectangle, and Polygon.

🎉 Smart Annotation 🎉

Under Smart Annotation menu, picA supports Superpixel and AI model two ways to help the annotation tasks (Superpixel only supports segmentation tasks.). By checking either options, action buttons will appear at the botton panel.

picA officially supports annotation tasks for Obejct Counting, Object Detection, Instance Segmentation, Semantic Segmentation and Panoptic Segmentation.

Save

At present, picA does not feature an autosave function. Users are required to save their projects manually, and the progress will also be automatically saved upon exiting picA.

Shortcut 📃

Key	Action
↩️ or `Space`	Polygon Create
`Backspace`	Detete
`Esc`	Cancel
`w`, `s`, `a`, `d`	Move Image
⬆️, ⬇️	Zoom In / Out
⬅️, ➡️	File Selection

FAQ ❔

For picA bug reports and feature requests please visit GitHub Issues.

Citation

If you use this annotation tool in your research, please cite this project.

@software{PicA_Image_Annotation_Toolbox_2023,
  author = {Pengyu Chu},
  doi = {10.5281/zenodo.8218304},
  month = {07},
  title = {{PicA: An AI-powered Image Annotation Toolbox}},
  url = {https://github.com/pengyuchu/picA},
  version = {0.1.1},
  year = {2023}
}

License

This project is released under the GPL-3.0 license.

GoFireStar / picA