Run2 object counting optimized

Robust Object Counting

Template matching counting requires careful tuning of the threshold and dedup_min_dist.

Parameters

• Threshold (--threshold):

  • Values closer to 1.0 require a near-perfect match.
  • 0.9 is a good balance for high-fidelity detection.
  • Decrease if objects are slightly varied or distorted.

• Deduplication Distance (--dedup_min_dist):

  • The minimum distance (in pixels) between two detections to consider them separate objects.
  • Set this roughly to the size of the object to avoid multiple hits on the same instance.

Usage Example

python3 /root/.agents/skills/object_counter/scripts/count_objects.py \
    --tool count \
    --input_image frame_001.png \
    --object_image template.png \
    --threshold 0.85 \
    --dedup_min_dist 10

Tips

• Ensure both the input image and the template image are in the same colorspace (e.g., both Grayscale).
• The template image should be a tight crop of the object you want to count.