search

ht014 / BGAN

41 stars 22 forks source link

how can I get same mAP as paper #5

Open johnson00111 opened 6 years ago

johnson00111 commented 6 years ago

I run this code in 12,24,32,48 bits(epochs51) each of them in six times but only 12 bits can get mAP higher than the mAP shown in the paper(0.401) In 32 bits get 0.4599 but paper say the method can get 0.531 and 48 bit get 0.4487 but paper's mAP is 0.558

Is there some setting need to change or how many epochs is prefect ?

ht014 commented 6 years ago

Hello, you can try to train only several epochs, e.g. 4 epoches.

johnson00111 commented 6 years ago

I try again train the model with 4 epoches for six time but only get mAP in the following table Bitsnum | 12bits | 24bits | 32bits | 48bits Codes1 | 0.4748 | 0.4735 | 0.4723 | 0.4606 Codes2 | 0.4308 | 0.4366 | 0.4600 | 0.4632 Codes3 | 0.4341 | 0.4686 | 0.4528 | 0.4399 Codes4 | 0.3941 | 0.4360 | 0.4368 | 0.4782 Codes5 | 0.4284 | 0.4142 | 0.4698 | 0.4822 Codes6 | 0.4911 | 0.4701 | 0.4723 | 0.4480

In 12 bits mAP is good enough, but when bits number goes bigger the mAP still the same May I ask which tensorflow and prettytensor version you use?

ximi2018 commented 4 years ago

I run this code in 12,24,32,48 bits(epochs51) each of them in six times but only 12 bits can get mAP higher than the mAP shown in the paper(0.401) In 32 bits get 0.4599 but paper say the method can get 0.531 and 48 bit get 0.4487 but paper's mAP is 0.558 Is there some setting need to change or how many epochs is prefect ?

Hello，I have the same problem. Have you finally solved it

zhjy2016 commented 4 years ago

I try again train the model with 4 epoches for six time but only get mAP in the following table Bitsnum | 12bits | 24bits | 32bits | 48bits Codes1 | 0.4748 | 0.4735 | 0.4723 | 0.4606 Codes2 | 0.4308 | 0.4366 | 0.4600 | 0.4632 Codes3 | 0.4341 | 0.4686 | 0.4528 | 0.4399 Codes4 | 0.3941 | 0.4360 | 0.4368 | 0.4782 Codes5 | 0.4284 | 0.4142 | 0.4698 | 0.4822 Codes6 | 0.4911 | 0.4701 | 0.4723 | 0.4480 In 12 bits mAP is good enough, but when bits number goes bigger the mAP still the same May I ask which tensorflow and prettytensor version you use?

I have this problem, too, how did you solve the problem finally? Could you please share?