thanks for your help
Opened this issue · 13 comments
Sorry to bother you. The pictures I verified are all assigned to the same category. What are your suggestions for defining the parameters? Thanks for your help.
@GAVANXU Could you tell me which version of keras and tensorflow to use, plz?
i got this error:
KeyError: "Can't open attribute (can't locate attribute: 'nb_layers')"
I am using version 2.1.5 of keras.and did not use the h5py module but directly loaded the model.
if weights_path:
model.load_weights(weights_path, by_name=True)
print('Model loaded.')
But the results of the final test are not ideal.
Of course, I'm not sure what's the problem with this. If you solve the trouble and teach me
@OeslleLucena i train own model the loss is 0.016 and acc is 0.98, But when testing the picture, all the photos were divided into 1 and the attack photos were divided into 0. Where did the problem occur? I only changed some hyperparameters.
@GAVANXU On which dataset did you train ? I am getting 1 for spoof and real photos with both the weights. Kindly help!!!
@GAVANXU Could u send me ur Code, and briefly describe the steps? thank u very mush.
mail id: gmyhbio@gmail.com
i train my own dateset. you can train the pretrain weight @saurav4098
@saurav4098 I am using the code given by the author only when loading the model with model.load_weights(weights_path, by_name=True) and not making other changes
@CacheTechLtd https://github.com/GAVANXU/anti-spoofing-keras
@GAVANXU 谢谢同胞[抱拳]
@GAVANXU Which data set do you measure？ What is the accuracy of the test?
@CacheTechLtd i tested the model with ReaplyATTACK and i found that is not accurate, is the same thing for you ??
@lotfiGipsyKing can you share ReaplyATTACK dataset with me? I cannot found it anywhere...