Hyperparameter for reproducing results in the paper

[MaginaDai]

I trained the model with the hyperparameters in PADing.yaml file but got a much worse result than the result reported in the paper.

I trained using pretrained_weight_panoptic.pth on the Panoptic data and get the result as follows:
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: Task: panoptic_seg
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: PQ,SQ,RQ,PQ_th,SQ_th,RQ_th,PQ_st,SQ_st,RQ_st,PQ_se,SQ_se,RQ_se,PQ_un,SQ_un,RQ_un
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: 34.0248,71.6245,40.5138,41.6926,77.9330,49.1312,22.4507,62.1021,27.5063,37.0349,74.3694,44.0742,8.4391,48.2921,10.2500
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: Task: bbox
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: AP,AP50,AP75,APs,APm,APl
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: 32.4348,47.4895,34.3758,14.6653,34.2537,49.9620
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: Task: segm
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: AP,AP50,AP75,APs,APm,APl
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: 30.9822,48.1893,32.5781,11.5876,33.2543,51.1710
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: Task: sem_seg
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: mIoU,fwIoU,mACC,pACC,unseen_IoU,seen_IoU,unseen_acc,seen_acc
[11/15 15:13:38] d2.evaluation.testing INFO: copypaste: 41.5571,36.6851,57.7758,51.8384,8.8335,45.4070,21.7921,62.0091

I suspect this might be due to the hyperparameter issues. Currently, there are several misalignments between the hyperparameters in PADing.yaml and the values stated in the paper. May I ask how to select the parameters?

Best

[Shuting He]

Thank you for your interest in our work!

You can refer to this issue to reproduce the result. #7