search

uta-smile / ASSUDA

Exploring Robustness of Unsupervised Domain Adaptation in Semantic Segmentation (ICCV 2021; Oral)
MIT License
12 stars 1 forks source link

batch size>1 #2

Open dongbo811 opened 2 years ago

dongbo811 commented 2 years ago

Hi I have a question. If I set the batch size greater than 1, and the code will report an error. However, the 2N mentioned in paper description does not seem to correspond.

viyjy commented 2 years ago

Hi, we have only tried batch size=1. What do you mean by "the 2N mentioned in paper description does not seem to correspond"?

dongbo811 commented 2 years ago

thx a lot for your answer. On this page, if we set the minibatch to N, we can get 2N samples. But, we found that it is not possible in the experiment.

At 2021-12-13 02:25:56, "viyjy" @.***> wrote:

Hi, we have only tried batch size=1. What do you mean by "the 2N mentioned in paper description does not seem to correspond"?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

viyjy commented 2 years ago

You are right, this code can only be applied to batch size=1, since we found that most of existing UDA methods in semantic segmentation use batch size=1. I can provide a new one which can handle batch size > 1. Before that, can you use batch size=1? Thanks.