Why is the division so chaotic?

There may be many reasons for edge confusion in image segmentation, some of which may be caused by noise, illumination changes and changes in the shape and size of objects in the image. In addition, if the traditional image segmentation algorithm is used, these factors may not be handled well, leading to edge confusion.

If deep learning algorithm is used for image segmentation, we can try to use better network architecture or better training data to improve the segmentation effect. For example, some studies show that using deeper networks can improve the segmentation accuracy.