search

vt-vl-lab / FGVC

[ECCV 2020] Flow-edge Guided Video Completion
Other
1.55k stars 263 forks source link

How can I train a new model on a different dataset? #59

Closed JHang2020 closed 2 years ago

JHang2020 commented 2 years ago

Hello, thanks for the excellent work. I want to test the performance of another dataset, so I may need to train a new model. Trainable parameters only exit in EdgeConnect and RAFT, right? So it means what I should do is to train EdgeConnect on the new dataset? Thanks for your help!

gaochen315 commented 2 years ago

What is this dataset about? In general, you do not need to re-train the models. But if your dataset is from a different domain, e.g., medical, you may have to re-train the models.

JHang2020 commented 2 years ago

Yean,it's exactly from a different domain. So I want to know how to train a new model. It seems that you don't give the reference for training. Thanks!

gaochen315 commented 2 years ago

For training RAFT, you can refer to the instruction here.

What if you directly run the code on your dataset? Is the result very bad?

JHang2020 commented 2 years ago

Thank you very much! I haven't run it directly, but I think it maybe difficult to work on a totally different dataset which is fine-grained. I will try it later.