search

zlmzju / part_reid

Code for ICCV2017 paper: Deeply-Learned Part-Aligned Representations for Person Re-Identification
123 stars 50 forks source link

reproduce results #8

Open yang502 opened 6 years ago

yang502 commented 6 years ago

I am try to reproduce the results on market1501. But the triplet loss is very small, nearly to zero. So after iter=35000+, the network did not update(loss=0). I can't reproduce the results(R@1:81%~,mAP:63%~), but get lower results (R@1:76%~,mAP:55%~).
Have you met this situation? What's the possible reasons ?

Anguliachao commented 6 years ago

@zlmzju hi ,I met the same problem. my model can't reproduce the 81% result on your market.caffemodel, instead is 76%. can you explain that? I followed the same instructions you posted on docs. Thank you