search

liliangqi / person_search

Joint Pedestrian Detection and Person Re-ID with PyTorch
90 stars 29 forks source link

Test accuracy #7

Open Dingyuan-Zheng opened 5 years ago

Dingyuan-Zheng commented 5 years ago

Thanks for the work! Is there anyone trained your own model using the provided code on the SYSU dataset and test your trained model? How much mAP accuracy can you achieve if you test the result with your trained model for the gallery size range from 50 to 4000?

chaste10ve commented 5 years ago

@Dingyuan-Zheng I have trained this model excluding the spatial transform part, unfortunately, the oim loss cannot convergence，and the mAP accuracy is only 24% at 10 epoch, also, the detection ap is under 10%. Can you give me some advice for that？

hh23333 commented 5 years ago

@Dingyuan-Zheng I have trained this model excluding the spatial transform part, unfortunately, the oim loss cannot convergence，and the mAP accuracy is only 24% at 10 epoch, also, the detection ap is under 10%. Can you give me some advice for that？

I also got a very low performance, have you found any reason of that? @chaste10ve

xuelimin commented 5 years ago

@Dingyuan-Zheng I have trained this model excluding the spatial transform part, unfortunately, the oim loss cannot convergence，and the mAP accuracy is only 24% at 10 epoch, also, the detection ap is under 10%. Can you give me some advice for that？

I also got the same result as yours, can you found the reasons for this?