search

lijin118 / LisGAN

LisGAN, Leveraging the Invariant Side of Generative Zero-Shot Learning, CVPR 2019
72 stars 21 forks source link

Can you provide the hyper-parameter to train on other datasets? #1

Closed scoobym closed 5 years ago

scoobym commented 5 years ago

Really appreciate your work, however, I use python lisgan.py --dataset AWA1 --nz 85 --attSize 85 --syn_num 300 --nepoch 350 to train on the AWA dataset and the results fluctuated between 0.5 and 0.6(zsl setting), maybe I'm using the wrong hyper-parameter. So could you kindly provide the hyper-parameter to train on other datasets, i.e., AWA, CUB, SUN

lijin118 commented 5 years ago

Please run the codes with pytorch 0.3.1 and you will get what you are expecting. You can refer to the updated readme.

scoobym commented 5 years ago

I run with the parameter you provide in readme, don't know why only got unseen class accuracy= tensor(0.6728) in (22/30)epoch

lijin118 commented 5 years ago

You mean 0.6728 or 0.6728%? If it is the former, I think it is normal. Otherwise, please check that you run the codes with pytorch 0.3.1.

scoobym commented 5 years ago

Thank you! Finally get the result !

naumanbashir70 commented 4 years ago

@scoobym I appreciate that you have run the code successfully. Could you help me in executing the code. It is giving me error 'train_loc' as this is not available in the provided dataset. How you manage to run this or you have you used other dataset?

scoobym commented 4 years ago

@naumanbashir70 The problem is that the dataset provided on the current website has changed and I used the original version of the dataset.

naumanbashir70 commented 4 years ago

@scoobym thanks for replying. Could you tell me where I can find original dataset? or could you send me if you have. Thanks

scoobym commented 4 years ago

@naumanbashir70 Sorry, I don't have the original dataset. You can ask author for it.

naumanbashir70 commented 4 years ago

okay, Thanks