This repository contains a Python Dash application for annotating keypoints in images.
- OpenCV
# Ubuntu/Debian
$ sudo apt update
$ sudo apt install libopencv-dev python3-opencv
# Other platforms
$ python3 -m pip install opencv-python --user
# Install pip dependencies
$ python3 -m pip install numpy dash dash_bootstrap_components --user
# Download repo
$ git clone https://github.com/luiscarlosgph/keypoint-annotation-tool.git
$ cd keypoint-annotation-tool
# Install package
$ python3 setup.py install --user
# You should be already inside the repo, where there is a sample 'data' folder to make this command work
$ python3 -m wat.run --data-dir data --port 1234 --maxtips 4
The --data-dir parameter should contain two folders: input and output.
The input folder should contain the images (*.jpg or *.png) to be annotated.
The output folder should be empty. The annotations will be stored there.
The --maxtips parameter sets the maximum number of tooltips that can be annotated per image.
By default this is set to four as there are typically two instruments in the scene with two tooltips
each (one per clasper).
Go to http://localhost:1234 with your favourite web browser.
When an image is annotated, the image file is moved from the input to the output folder.
Each image in the output folder (e.g. example.png) is accompanied by two annotation files:
-
example_seg.png: single-channel PNG image. All pixels are set to zero except those that have been clicked during the annotation process. The first keypoint clicked is annotated with intensity 1, then second with intensity 2, and so on. -
demo.json: contains a list of annotations in JSON format, for example, if four tooltips are clicked, it will contain something like{"tooltips": [{"x": 263, "y": 168}, {"x": 586, "y": 301}, {"x": 581, "y": 210}, {"x": 500, "y": 116}]}
The script read_annotations.py is provided as a staring point to process your annotated images.
# Run this command after annotating the sample image provided with the repo
$ python3 -m wat.read_annotations --dir data/output
This code is distributed under the MIT license.