Skip to content

Latest commit

 

History

History
40 lines (31 loc) · 1.61 KB

README.md

File metadata and controls

40 lines (31 loc) · 1.61 KB

Label the images

Requirements:

  • python 3.6
  • PyQt5
  • requests
  • skimage

Install python 3.6 and pip, and install the requirements:

pip install python-qt5 requests scikit-image #for windows/mac 
pip install --user pyqt5 requests scikit-image #for ubuntu

Then, run the main.py to start:

python main.py

Steps:

  • Download this project and install the requirements.
  • Select or input the beginning image ID in the top "Image ID" box.
  • Input the key words (split by newline character)for two images in their corresponding text areas or select the check boxes.
  • Click 'Next Sample' to continue, and the result will be automatically saved in './caption_result/' dir.

Relus to follow：

  • The key words mainly depend on your subjective judgement, the candidates in check boxes are only for reference purpose.
  • If you think the candidate labels in check boxes match the image content, just tick it.
  • As for the key words matching the image but missed in check boxes, you should manually type them.
  • Each image should have 3-5 labels

Example

image

The program may be stuck when you press the button, this is caused by the remote image requesting, please wait a moment ...

The "Category" and "Attribution" show some classification information and attributions, they are predicted by the Place365 model and are only for reference purpose, because they are not reliable enough.

When your work finished, pack the './caption_result/' to file 'start-id_end-id.zip', e.g. '10_100.zip', and upload it in this site

Thank you very much!