results are different from pycocotools

[Shawnnnnn]

I have tried to use same gt and predictions to caculate mAP with this script and pycocotools, then I got different results, why? The mAP with this script is lower than pycocotools, it seems FP is high and they are from those bboxs with very low scores. Is pycocotools filtering a lot of bboxes with very low scores? How can I get same mAP from this script? Thanks for your work！

[ashok-arjun]

Did you check the IoU threshold for both the scripts?