search

PKU-ICST-MIPL / OPAM_TIP2018

Source code of our TIP 2018 paper "Object-Part Attention Model for Fine-grained Image Classification"
91 stars 32 forks source link

about part alignment #11

Open ch0511 opened 4 years ago

ch0511 commented 4 years ago

When training, is ClassNet fine-tuned by 2 aligned parts together to get one PartNet, or fined-tuned seperately by 2 parts to get Part1Net & Part2Net?

If 2 aligned parts are used to fine-tune the ClassNet together, then why is alignment necessary?

I just can't think through these questions, could you please help me? thank you!^_^

mumuhe commented 4 years ago

When training, is ClassNet fine-tuned by 2 aligned parts together to get one PartNet, or fined-tuned seperately by 2 parts to get Part1Net & Part2Net?

If 2 aligned parts are used to fine-tune the ClassNet together, then why is alignment necessary?

I just can't think through these questions, could you please help me? thank you!^_^

@ch0511 Yes, 2 parts are used together to get the PartNet. The alignment is done after getting PartNet by Part Alignment.

ch0511 commented 4 years ago

Thank you!