about Metrics

[laugher07]

i retrain the model, Why did I obtain metrics close to those reported in the paper on the THuman2 dataset, but the metrics on the CAPE dataset are not as good?
like cape-nfp(chamfer: 1.1177， p2s:0.9627, nc: 0.0520) , cape-fp(chamfer:0.8776, p2s:0.8042, nc:0.0442)

[zjh21]

My retrained model has worse metrics on both THuman2.0 and CAPE. It's like:
{'cape-easy-NC': 0.04301762208342552,
'cape-easy-chamfer': 0.9157772064208984,
'cape-easy-execution_time': 0.44610512733459473,
'cape-easy-p2s': 0.8151900768280029,
'cape-hard-NC': 0.050745654851198196,
'cape-hard-chamfer': 1.1736739873886108,
'cape-hard-execution_time': 0.45782910426457724,
'cape-hard-p2s': 1.015979290008545}
I also encounter another strange phenomenon that training takes actually a few hours on a 4090, rather than the reported 2 days on a 3090. I am using THuman2.0 for training.