As a [stakeholder], I want to [description of the desired outcome].

**Description**
When one box completely covers another box, you can't add feedback to the encapsulated box.

**Bug Summary:**
Box # 3 (which is a hallucination in this case but could potentially be a real seed if one seed was very long and oriented at an angle) completely encapsulates box # 1. #1 is correctly ID'd (though the ID name contains "D" for some reason, maybe because it is a dimorphic seed), but I can't "reach" it to confirm the ID.

**Steps to Reproduce:**
Load the base image at the bottom of this post in Nachet
Classify with 27 species model

**Expected Behavior:**
Two seeds would be detected and classified.

**Actual Behavior:**
A third box (labeled # 3 in the image below) covers a very large area, completely encapsulating the seed labeled # 1. Although I can provide the feedback that # 3 is not a seed, it does not remove the box and I can't click on box # 1 to provide feedback on that seed.

**OS Version:** Windows 11

**Browser Version:** Edge

**Logs and Screenshots:**



**Browser Console Logs:**
[Include relevant browser console logs, if applicable]

**Extra Screenshots:**
[Attach any relevant extra screenshots]

**Additional Context:**
[Include any other context about the problem here.]

**Suggested Fix:** (facultatif)
The phantom box disappearing once it was marked as "not a seed" would solve this particular issue, but if both were real seeds that wouldn't work. In that case, maybe a way to tab through the seeds or select from the seed list in the results section might work?

Reproduce the overlap in Nachet by loading the linked base image and classifying it with the 27 species model. Start at the classification results view and inspect how overlapping detection boxes are selected or removed. Done means a covered seed can still be selected and given feedback, including when both overlapping boxes are valid seeds.

Escrita pelo modelo de indexação a partir do texto da issue.