Why can't I get the mAPs reported in the paper?

[SquirrelAtHW]

Trained model with default setting (default data separation, config: configs/PseCo/PseCo_faster_rcnn_r50_caffe_fpn_coco_180k.py) and I am getting mAP 19.6 for 1% labeled data and mAP 31.4 for 10% labeled data, against 22.43 for 1% and 36.06 for 10% reported in paper.

Also for each iteration, I keep getting "unsup_precision: 0.0000, unsup_recall: 0.0000". Does this look suspicious?

What could be the potential problems? Followed are some of my guesses:

1. Incompatible environment, which might mistaken some calculations
2. Non-optimal configs. In this case, may someone share an optimal configuration?

[miaowumonsyer]

Me too,for each iteration, I keep getting "unsup_precision: 0.0000, unsup_recall: 0.0000"