Question about VisDA2017

[miss-rain]

Good work!

we know VisDA2017 has three part: train dataset, validation dataset, and test dataset.

In code (train_src_visda.py and train_tar_visda.py)

for training source model, G-SFDA use VisDA's train dataset (90% train and 10% test)
for training target model, G-SFDA use VisDA's validation dataset (shuffle =ture, batchSize)
for testing target model, G-SFDA also use VisDA's validation dataset (shuffle =false, batchSize*3)

one question: why not use VisDA's test dataset?

[Albert0147]

Hi, as the VisDA does not directly provide the label for its 'test' set, we do not use it, as all recent DA papers did.

[miss-rain]

Thanks for relpy.

another question: why set batchSize*3 when testing target model?

G-SFDA ues all and same VisDA's validation dataset to train and test target model, is it right?

[Albert0147]

Just to accelerate the evaluation, since you do not train this time.

[miss-rain]

Thanks but sorry, i don't understand it

[Albert0147]

It does not matter, it will not influence the results. Small batch size means you need to load more times from the dataset to gpu memory and may slow the evaluation.

[miss-rain]

Ok got it.

G-SFDA use all and same VisDA's validation dataset to train and test target model, is it ok?

[Albert0147]

Yeah, just as all DA methods do.

[miss-rain]

When training source model ,
G-SFDA use 90% VisDA's train dataset for traing ,and use 10% VisDA's train dataset for testing.

while when training target model,
G-SFDA use 100% VisDA's validation dataset for traing ,and use same 100% VisDA's validation dataset for testing.

Did i understand right?

[Albert0147]

yeah

[miss-rain]

Nice.

in code train_src_visda.py, lines 252.

When training source model ,
Why G-SFDA use 100% VisDA's validation dataset for testing source model?

[Albert0147]

just test on target domain...

[miss-rain]

That's mean test init target model?

[Albert0147]

yep