search

michuanhaohao / keras_reid

Margin Sample Mining Loss: A Deep Learning Based Method for Person Re-identification
188 stars 59 forks source link

identity_num question #9

Closed XrsSjtu closed 5 years ago

XrsSjtu commented 5 years ago

Hello,why does the identity_num need to be 6273 when i load your pre-train weights?

michuanhaohao commented 5 years ago

@XrsSjtu Because I use multi datasets to pre-train the model, your can remove the classifier FC layer.