search

mengrang / DFL-CNN-tensorflow

TF-Implementation of "Learning a Discriminative Filter Bank within a CNN for Fine-grained Recognition"
74 stars 17 forks source link

About Accuracy #1

Open Ien001 opened 6 years ago

Ien001 commented 6 years ago

hi, I`m working on reimplementing "Learning a Discriminative Filter Bank Within a CNN for Fine-Grained Recognition" now, but I cannot get the result in the paper, do you get the same accuracy listed in the paper?

Ian

mengrang commented 5 years ago

Do you implement the layers nonrandomly initiallation?it makes the key role in the accuracy

Ien001 commented 5 years ago

Yeah.. but it seems like I'm failed on it, the accuracy is no different after I finish nonrandomly initiallation. Many ppl said that there are some tricks in the paper😳, I don't know, maybe I did sth wrong.

mengrang commented 5 years ago

Thanks for your comments, I am also intersted in the initialltion works.You can click here for My understanding of the initiallztion.

fxle commented 5 years ago

@Ien001 How do you try to nonrandomly initiallation?Can you share you idea? Thank you

fxle commented 5 years ago

@mengrang How accurate are you getting? In addition，Where to get these two files?'AgriculturalDisease_train_annotations.json'',"AgriculturalDisease_validation_annotations.json"

mengrang commented 5 years ago

The acc on my dataset is just OK, the two json files is my dataset

Ien001 commented 5 years ago

Hi, I have read it, @mengrang it seems okay, have u implement ur idea?

doublemanyu commented 5 years ago

Have you implement the non-random initialization? @mengrang @Ien001 @fxle

chaerlo commented 5 years ago

@Ien001 what‘s the accuracy u implement the non-random initialization？