Skip to content

A lightweight image annotation framework for deep-learning-based object detection, built using pure OpenCV.

License

Notifications You must be signed in to change notification settings

amogh7joshi/annotator

Repository files navigation

Annotator

Annotator is a lightweight framework for constructing object detection annotations. It's built only using OpenCV, with no other third-party libraries involved. The primary frontend consists of a makeshift GUI with buttons corresponding to different labels and a two-click bounding-box creation interface, and the backend parses the annotations and places the bounding box coordinates into text files corresponding to the images.

Installation

To use Annotator yourself, first clone this repository:

git clone https://github.com/amogh7joshi/annotator.git

Then, enter the directory and execute the following command to build the compiled program, which will be triggered by the build.sh build script:

chmod +x ./build.sh
./build.sh

Now, you can edit the config.txt file with your own parameters for execution. Each of the three lines corresponds to the following parameters:

  1. The path to the directory containing images (or a directory containing directories of images).
  2. The list of labels that you want to annotate, space-separated.
  3. Whether to recursively search through the sub-directories of the parent path (either 'true' or 'false'.)

Finally, execute the following command and an annotator session will begin:

./annotator

Usage

After the Annotator window launches, it will sequentially load each of the images in the directory path that you have provided. From here, you will be presented with a window similar to the following image:

Base Image

You can then use the two-click interface by clicking on a specific point of the image (excluding the buttons) and then clicking on a second point. For each set of two clicks, a bounding box will appear. This will then be stored in a internal vector of bounding boxes alongside the selected label, and the process will repeat for each bounding box on each individual image.

Annotated Image

You can click on the different label buttons at the top to switch between labels, and this will correspondingly be recorded internally and will appear as such in the output annotation text files.

Additionally, the following keyboard shortcuts are part of Annotator:

  1. q: Exit the session and close all windows.
  2. c: Clear the annotations for the current image.

License and Contributions

GitHub

The code in this repository is licensed under the Apache 2.0 License, see LICENSE for more details and terms of usage. If you have any contributions that you feel may improve Annotator, feel free to contact me at joshi.amoghn@gmail.com, and if you notice any pressing issues, don't hesitate to head to the Issues portion of the repository and let me know.

About

A lightweight image annotation framework for deep-learning-based object detection, built using pure OpenCV.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published