Calculator Input
Enter one box per line using x1,y1,x2,y2,score,label.
Example Data Table
| ID | x1 | y1 | x2 | y2 | Score | Label | Notes |
|---|---|---|---|---|---|---|---|
| 1 | 10 | 12 | 90 | 110 | 0.95 | person | Likely retained because it has the highest score. |
| 2 | 14 | 18 | 88 | 108 | 0.89 | person | Likely suppressed because it overlaps strongly with box 1. |
| 3 | 120 | 20 | 190 | 110 | 0.90 | person | Likely retained because overlap with box 1 is small. |
| 4 | 18 | 130 | 95 | 220 | 0.81 | dog | Usually retained if class-aware mode is enabled. |
| 5 | 24 | 138 | 98 | 225 | 0.73 | dog | Often suppressed by box 4 when IoU is high. |
| 6 | 125 | 135 | 210 | 225 | 0.67 | cat | Usually survives if it does not exceed the threshold. |
Formula Used
Intersection over Union (IoU) measures overlap between two boxes.
NMS procedure: sort boxes by score, keep the highest box, compare it against remaining boxes, suppress boxes whose IoU exceeds the threshold, and repeat until no candidates remain or the detection cap is reached.
How to Use This Calculator
- Enter each detection as x1,y1,x2,y2,score,label.
- Set the IoU threshold for overlap removal sensitivity.
- Set the score threshold to remove weak detections first.
- Choose whether suppression should be class-aware.
- Set the maximum number of detections to keep.
- Run the calculator to view kept boxes, suppression details, and the graph.
- Use the export buttons to save results as CSV or PDF.
Frequently Asked Questions
1. What does Non Max Suppression do?
It removes duplicate detections that point to the same object. The algorithm keeps the strongest box and suppresses weaker boxes whose overlap exceeds the chosen IoU threshold.
2. What is a good IoU threshold?
Common values range from 0.3 to 0.6. Lower values suppress more aggressively, while higher values allow more nearby boxes to remain.
3. Why use a score threshold first?
Score filtering removes weak detections before suppression starts. That reduces noise, shortens computation, and makes the retained boxes easier to interpret.
4. What is class-aware NMS?
Class-aware mode suppresses boxes only when their labels match. That prevents a strong person box from suppressing a nearby dog or car detection.
5. Can I use normalized coordinates?
Yes. You can use pixels or normalized values, but every box must use the same coordinate scale for the IoU calculation to remain valid.
6. Why is a lower-score box sometimes kept?
A lower-score box survives when its overlap with stronger boxes stays below the threshold, or when class-aware mode prevents comparison across different labels.
7. What happens when max detections is small?
The calculator stops keeping new boxes once the limit is reached. Remaining candidates are listed as dropped by the final cap, even if they were not suppressed by IoU.
8. Is this suitable for model debugging?
Yes. It is useful for inspecting detection overlap, threshold effects, retained outputs, and class-aware behavior during evaluation, demos, or pipeline tuning.