search

GNAYUOHZ / ReID-MGN

Simple pytorch unofficial implement of paper: Learning Discriminative Features with Multiple Granularities for Person Re-Identification
235 stars 61 forks source link

How can I classify 200,000 categories? use softmax loss . #13

Closed xf739645524 closed 5 years ago

GNAYUOHZ commented 5 years ago

If I understand correctly，change num_classes in network.py to 200,000. Maybe I should write it in argparse.

xf739645524 commented 5 years ago

Okey ， I trained for a week and the crossentropy loss finally came down.

guanyonglai commented 4 years ago

Hi, how about the results when you train 20W IDs? Better than less IDs?

xf739645524 commented 4 years ago

Hi, how about the results when you train 20W IDs? Better than less IDs?

This loss is very difficult to drop

guanyonglai commented 4 years ago

This loss is very difficult to drop

I have trained on my own dataset, The 8000ID training produced the best results,If you add more ID, the accuracy starts to go down.Have you solved this problem yet?