search

VisionLearningGroup / visda-2018-public

45 stars 34 forks source link

[open set domain adaptation] Use test image when training #12

Open lianqing11 opened 6 years ago

lianqing11 commented 6 years ago

Can we use test image when training the domain adaptation model?

xcpeng commented 6 years ago

Yes, you can use the test images

Regards,

Xingchao Peng, Ph.D. Student Image and Video Computing Group, Boston University

On Wed, Aug 1, 2018 at 3:08 AM, LianQing notifications@github.com wrote:

Can we use test image when training the domain adaptation model?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/VisionLearningGroup/visda-2018-public/issues/12, or mute the thread https://github.com/notifications/unsubscribe-auth/AHgmrhBKeg5h7LGQc9nX3EQFvr1zSYfJks5uMVPagaJpZM4Vp9YK .

lianqing11 commented 6 years ago

Thanks for your respond.

I am curious why choose this kind of setting that at the final stage we can use all of the target image to test our model instead of splitting the target image to train and test set and only can use the train set when training. Because at a real scenario, we can't get the test image when we train the model.

xcpeng commented 6 years ago

Hi, Lianqing

That's really a good idea. I would say as most of the Unsupervised Domain Adaptation in the research community use this setting, i.e. training data with labels and testing data without labels, we follow this schema.

But we will take consideration of your suggestions in the future. Thanks

Regards,

Xingchao Peng, Ph.D. Student Image and Video Computing Group, Boston University

On Thu, Aug 2, 2018 at 12:29 AM, LianQing notifications@github.com wrote:

Thanks for your respond.

I am curious why choose this kind of setting that at the final stage we can use all of the target image to test our model instead of splitting the target image to train and test set and only can use the train set when training. Because at a real scenario, we can't get the test image when we train the model.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/VisionLearningGroup/visda-2018-public/issues/12#issuecomment-409802166, or mute the thread https://github.com/notifications/unsubscribe-auth/AHgmrj6JdYNEYLibi4V127K74Ra7Y0DOks5uMoAngaJpZM4Vp9YK .