search

huanghoujing / beyond-part-models

PCB of paper: Beyond Part Models: Person Retrieval with Refined Part Pooling, using Pytorch
331 stars 81 forks source link

About the performance of model trained on market1501 #35

Open wangyingquan opened 4 years ago

wangyingquan commented 4 years ago

Thanks for your work, I'm sorry to bother you. I try to train the PCB on marker1501 , but the performance I get is very different from yours ( cmc1=76.69% cmc5=85.72%).At the same time , I used the 'model_weight.pth' you provided to test the models can reach the performance about cmc1=93.02% .Do you know the reason?

huanghoujing commented 4 years ago

Hi, thanks for your attention. Do you use Python 2.7 and Pytorch 0.3?

wangyingquan commented 4 years ago

Yes.But I forgot that I modified a line of code . Now the performance of model is almost same as you show . Thank you for your reply !