PointTinyBenchmark
PointTinyBenchmark copied to clipboard
Cannot reproduce the upper bound mAP of P2PNet.
Dear all, Thanks for the awesome work.
I use the config file 'p2p_r50_fpns4_1x_fl_sl1_coco.py' to train P2PNet with the center annotations. But only 50.25 mAP is achieved. What should I do to reproduce 57.47 mAP reported in your paper?