Establishing the ground truth for difficult cases when it comes to images


I need to create some labels with labelImg for specific objects, but there are two main factors, which make the process of creating bounding boxes a difficult task.

  • The objects of interest are small compared to the size of the image and stacked together, think like two or more rectangles stacked together.
  • The quality of the image is not very good, which means that if I zoom in a very high level, the quality of the image worsens, thus I am not even able to distinguish their boundaries.

Has anyone encountered a similar problem?


Posted 2017-07-10T11:54:00.393

Reputation: 153

No answers