Tuning of yolov4 model

Hi, i have currently trained a custom yolov4 model (TLT ver 3.0), and when I used the trained model to perform inference on a test data, i realized that there are a lot of false positive ( the model is able to detect the target, but there are multiple bbox over the target itself) . I do know that there is a Nms configuration for yolov4 model, and a possible way to remove those bbox is to set a higher confidence level threshold. However for the cluster_iou_threshold parameter, should i increase the value for it or decrease?

Yes, you can change the threshold. See more details in YOLOv4 — TAO Toolkit 3.22.05 documentation.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.