Data Labeling

[yehyunsuh]

What

Data Labeling

Why

To start training in the future

TODO

• [ ] Using Labelme
• [ ] Using Tablet

[yehyunsuh]

For Windows,

• Download anaconda
• Install labelme

  conda create --name=labelme python=3
  conda activate labelme
  pip install labelme

• Activate labelme

  labelme .\resized_jpg\ --labels .\labels.txt

[yehyunsuh]

Labelme is based on clicks, and creates polygon from it. Really hard to annotate pores. Looking at different options.

• https://dataloop.ai/docs/create-semantic-segmentation Looks really good but not free.
• https://github.com/heartexlabs/awesome-data-labeling
• https://mangastorytelling.tistory.com/entry/Open-Source-%EB%8D%B0%EC%9D%B4%ED%84%B0-%EB%9D%BC%EB%B2%A8%EB%A7%81-%ED%88%B4-%EB%AA%A9%EB%A1%9D-List-of-open-source-annotation-tools-for-ML Currently trying these two options
• https://github.com/opencv/cvat My computer cannot hold on to the server.
• https://github.com/abreheret/PixelAnnotationTool

[yoonjongyeon]

• Tried labeling via handwriting app 'GoodNote' but had issue of image's resolution getting small
• Used image editing app 'ProCreate' instead -> problem that showed in 'GoodNote' was solved -labeled nano particle in blue & pore in red

[yoonjongyeon]

• Made code for getting label from labeled images

• Some minor problems -unlabeled part in nano particle (pore) -> filled that part by erosion and dilation -when saving array to image, value of pixels changed -> saved array in to pickel

final result (labeled image of nano particle)