pose-annotation-tool for dont be turtle proj

About

This is an annotation tool for the dont be turtle proj.

Installation

Git Clone and Cmake

$ git clone https://github.com/motlabs/dont-be-turtle-pose-annotation-tool 
$ cmake CMakeLists.txt
$ make

How to Use

1) Placing your dataset

Placing your dataset at ./images_for_annoation by creating a subfolder. For example, when you place the LSP dataset, the images could be located at

$IMAGE_DIR=./images_for_annotation/lsp_dataset/images/

2) Running the program

$ python run_annotation.py --imagedir=$IMAGE_DIR

3) Starting Annotation

When running this program, two windows are popped up:

• One shows the images that is currently annotating.
• Another shows which keypoint is currently pointed.

We can have three choices before starting the annotation:

- Press ENTER to annotate the current image
- Press SPACE to skip the current image

Then the annotation is manipulated by the following:

- Left Click  + Any key: Annotate a visible joint and register
- Right Click + Any key: Annotate a occluded joint and register
- TAB: Current joint is not applicable for current image

4) Getting Results

The annotation results are formatted by a JSON format. For example,

{
	 "image_path": "./images_for_annotation/lsp_dataset/images/front_normal_10754.jpg",
	 "head": [ 248 ,127, 0 ],
	 "nose": [ 251 ,284, 0 ],
	 "Rshoulder": [ 87 ,406, 0 ],
	 "Lshoulder": [ 412 ,399, 0 ]
}

And, being recorded at

$LABEL_DIR=label_annotated/lsp_dataset/labels

Note that one label json file is generated for one input image (one to one corresponding relation for the image and label).

Code Reference

• https://github.com/suriyasingh/pose-annotation-tool

Feedback

• Jaewook Kang (jwkang@gmail.com)