Investigating NMS parameters for improved mAP results

[jxmmy7777]

Hello @stepankonev

Thank you for sharing your code on GitHub. I have questions regarding the non-maximum suppression (NMS) algorithm as used in Technical Report 3.53, and I'm wondering if you could provide more details on the parameters you chose for NMS? Specifically, I've been training the model with the final_RoP_Cov_Single.yaml configuration file and I've noticed that while I'm getting similar results without NMS for metrics like minADE, minFDE, and miss rate (in #27 (comment)), the mAP metric doesn't match up (0.32 v.s. 0.22 for vehicle in 8s). Any additional information you could provide would be greatly appreciated. Thank you.

Best regards,