General image labeling advise

I am new to ML, and roboflow. i want to label images of honeybee comb elements so that a future app can be used to identify (statistically summarize) what the beekeeper sees during a hive inspection. I’m not sure which marking feature I should use to distinguish (train) between eggs and larva. Should I mark the whole hexagon cell, use a box around the main component (egg or larva) or use the detailed polygon with multiple anchors to tie is as close to the image shape as possible (hope this makes sense)? I’m attempting to attach are an example.

Here’s an example of bee larva (they rarely fill the whole cell however).

Hi @uclatico - my advice is to label both the hexagon and the entity within it. You’d be able to use post-processing to count what % of the hexagons are occupied.

For these types of images, bounding boxes are fine. I would just make sure they are representative of the data your model will see in the wild.

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.