Hello,

I'm concerned about the quality of the annotations in this dataset. It appears that automatic annotation tools may have been used, as the annotation quality is quite poor. The main challenge in this competition seems to be detecting Trophozoites, not because they are relatively small objects, but because they look very similar to other unlabeled (background) objects in many images.

I'm not familiar with what a Trophozoite is, but a general rule in computer vision: if it's difficult to detect objects with the human eye, it will be even more challenging for a model to do so. As you can see in the attached images, Trophozoites look very similar to non-annotated objects in the background.

My main questions are:

1. Are you certain that no objects were missed during the annotation process?
2. Is it possible for us to re-annotate the data to improve its quality? (Not sure if this will be helpful (: because still don't know what Trophozoite looks like)