search

Albert0147 / AaD_SFDA

Code for our NeurIPS 2022 (spotlight) paper 'Attracting and Dispersing: A Simple Approach for Source-free Domain Adaptation'
61 stars 6 forks source link

Low accuracy of OfficeHome dataset #1

Open TL-Wucw opened 1 year ago

TL-Wucw commented 1 year ago

Thanks to the author for contributing the code! Based on the hyperparameters provided in the paper, I found in my experiments that the accuracy of the OfficeHome dataset is very low (only 1.58% in task A->P) and seems difficult to fit.

The source domain model parameters I used were provided by SHOT.

Albert0147 commented 1 year ago

Hi. Please do not use other source model. Do use pytorch 1.3 with cuda 10.0 for training with our code. There is definitely something wrong, considering even source model can have better performance.

TL-Wucw commented 1 year ago

Thanks for your reply! Can you provide the following hyperparameters for the source model training on the Office31 and OfficeHome datasets?

Albert0147 commented 1 year ago

Hi, I upload the unclean code for office-home and office31, I think the hyperparameters are the same in the paper.

TL-Wucw commented 1 year ago

Thanks!