search

truongthanhdat / ADDA

Adversarial Discriminative Domain Adaptation
8 stars 2 forks source link

test on office dataset #1

Open xh135345 opened 6 years ago

xh135345 commented 6 years ago

I run train_classifier.py to train a base resnet50 model on office-dslr dataset,but the accuracy of webcam from test_classifier.py is low,about 20%, and the accuracy of validation set of dslr is about 60%, how i should preprocess these dataset?

truongthanhdat commented 6 years ago

@xh135345 I think you should train adaptative (train_adapt.py) from office-dslr dataset to webcam dataset before you test on webcam dataset.

xh135345 commented 6 years ago

Thanks for your reply. But I just consider the baselineļ¼Œsource only. I think the accuracy of webcam should be close to the source's before i train adaptation, and my view is based on some papers.