A question about training

[junjie1003]

Could you please tell me the difference between the first “config.pre_train=True” and the later “config.pre_train=False”? Thank you!

[MehmetAygun]

True one doesnt use instance loss for training while false one uses all of them. In my experiments, training firstly backbone then finetuning improved results.

[junjie1003]

Mehmet Aygün ***@***.***>于2021年7月19日 周一14:40写道：

True one doesnt use instance loss for training while false one uses all of them. In my experiments, training firstly backbone then finetuning improved results. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANRGHJGWEDNKLSUGXQ73QX3TYPCEZANCNFSM5ASYWBBA> .

Thank you for your response. And I still have a question. Do you finetune the whole network? Or only finetune four decoder after the encoder-decoder architecture?-- Junjie Yang, College of Intelligence and Computing, Tianjin University

[MehmetAygun]

I finetuned all of them. The false option doing that.

[junjie1003]

Mehmet Aygün ***@***.***>于2021年7月20日 周二00:54写道：

I finetuned all of them. The false option doing that. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANRGHJCATSDWPU5SHSJLA33TYRKFVANCNFSM5ASYWBBA> .

Thank you very much!-- Junjie Yang, College of Intelligence and Computing, Tianjin University