search

thuml / Universal-Domain-Adaptation

Code release for Universal Domain Adaptation(CVPR 2019)
270 stars 52 forks source link

Pretraining on ImageNet #9

Open tarun005 opened 3 years ago

tarun005 commented 3 years ago

You had mentioned that the backbone network is ResNet-50 pretrained on Imagenet.

https://github.com/thuml/Universal-Domain-Adaptation/blob/5d7caa95af7e3675305c542253c4e372801897d2/net.py#L37

But in many experiments in the paper, the labels in the unsupervised target overlap with that of the supervised source labels in the ImageNet. Is it justified that you pretrain on supervised ImageNet labels?

youkaichao commented 3 years ago

That's a nice point. We use pre-trained models because they are so popular and are used by default. Maybe you can investigate how pre-trained models affect the performance :)