google-research
The `response_seq_class_m` is used for the input sequence (why do the code randomly change the label of the Input Sequence),
Why do I randomly change the label of the Input Sequence? I don't really understand.
label_mapping = {'none': new_label, 'rand_cls': rand_cls, 'real_n_fake_cls': real_n_fake_cls, 'rand_n_fake_cls': rand_n_fake_cls, 'real_n_rand_n_fake_cls': real_n_rand_n_fake_cls}
What are the benefits of ‘’rand_n_fake_cls‘’ over 'none'？
this is mainly to avoid shortcuts in learning. For example, once the model confirms a fake/noise object appears, all the rest objects can be easily recognized as fake/noise.
On Sat, Sep 24, 2022 at 10:27 AM huimlight @.***> wrote:
label_mapping = {'none': new_label, 'rand_cls': rand_cls, 'real_n_fake_cls': real_n_fake_cls, 'rand_n_fake_cls': rand_n_fake_cls, 'real_n_rand_n_fake_cls': real_n_rand_n_fake_cls}
What are the benefits of ‘’rand_n_fake_cls‘’ over 'none'？
— Reply to this email directly, view it on GitHub https://github.com/google-research/pix2seq/issues/18#issuecomment-1257023884, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAKERUJHEWRWRRNSCWNZMHDV742ZFANCNFSM6AAAAAAQUXFSI4 . You are receiving this because you are subscribed to this thread.Message ID: @.***>
