search

yxgeee / SpCL

[NeurIPS-2020] Self-paced Contrastive Learning with Hybrid Memory for Domain Adaptive Object Re-ID.
https://yxgeee.github.io/projects/spcl
MIT License
316 stars 66 forks source link

problem with rerank #37

Closed Git-Yxt closed 3 years ago

Git-Yxt commented 3 years ago

i use spcl_train_usl.py on duke with 1 V100,I got the result mAP=64.8% R1=79.6% R5=88.9% R10=91.1%.When I set rerank=True ,I got the result mAP=78.2%,R1=83.5%,R5=89.4%,R10=91.8%.I think it is a little strange that mAP increases a lot while R1,R2,R3 increase only a little.