How to reproduce your best result?

[Xiangyu-CAS]

Hi, I was trying to reproduce your best result (mAP=88.7%) by simply runing

python tools/train.py --config_file='configs/softmax_ranked.yml' DATASETS.NAMES "('market1501')" 

but I got a much worse result at mAP=86.5%, with 2% margin to claimed result. Moreover, it's even lower than using triplet loss. What I can do to get your claimed result?

model	mAP	loss
r50-ibn-a(my result)	87.1%	softmax + triplet
r50-ibn-a(my result)	86.5%	softmax + ranked
r50-ibn-a(your result)	88.7%	softmax + ranked

Thank you!

[Qidian213]

Did you change training paramters ?
There are some exp result, you can refer (Resnet50_ibn_a).
Try more epoch ?

`

w	r	T	DukeMTMC-ReID	Market1501
mAP	rank1	mAP	rank1
0.2	0.7	1.0	80.2	90.1
0.4	0.7	1.0	80.8	90.9
0.6	0.7	1.0	80.6	90.2

w	r	T	DukeMTMC-ReID	Market1501
mAP	rank1	mAP	rank1
0.4	0.6	1.0	80.3	90.6
0.4	0.7	1.0	80.8	90.9
0.4	0.8	1.0	80.8	90.4

w	r	T	DukeMTMC-ReID	Market1501
mAP	rank1	mAP	rank1
0.4	0.7	0.5	80.4	90.6
0.4	0.7	1.0	80.8	90.9
0.4	0.7	5.0	80.3	90.6

`

[say2sarwar]

I used your RankedListLoss for car196 dateset but i can't reproduce the results. the loss is increasing with every epoch. any thoughts?

[Mr-Da-Yang]

Hello, what is the meaning of these hyperparameters？
Looking forward to your reply！thank you