Can't reproduce the result of lemniscate model

[boom85423]

Hello, I think your code of AND is nice.

I already reproduce the result of random model(85.7%) but lemniscate model(85.96%) which is different to 86.3 in paper. Basically, I follow the original code of lemniscate, please inform me if there are some hyper-parameter or something I miss.

Thank you~~~

[boom85423]

@Raymond-sci I need your help~~~

[Raymond-sci]

thanks but such a minor difference is possibly experimental bias. Repeated trials may produce better results even than the reported ones. as for parameters, can you please try to set the resize=32 in config file config/base.yaml to resize=48. turning on the random_horizontal_flip may also help a bit.

[boom85423]

Do you mean use in resize=48 and random_horizontal_flip=True in original lemniscate ? But the accuracy is drop to 73%...

[Raymond-sci]

ok then let me check, will let you know if there is any problem found