The accuracy rate of DANN published in this paper in Office31 data set is 82.2%, but the actual result I get by running this DANN code is much higher than 82.2%. Is this normal? It was also higher on the OfficeHome data set than published in the paper.
We report both the results of the original paper and the reproduced results. For the image classification task, the results are here. For DANN, our implementation does achieve better results on both the Office31 and OfficeHome datasets.
The accuracy rate of DANN published in this paper in Office31 data set is 82.2%, but the actual result I get by running this DANN code is much higher than 82.2%. Is this normal? It was also higher on the OfficeHome data set than published in the paper.