Run run.sh to reproduce the files in logs/
Example evaluation:
python3 main.py /gpfs01/bethge/data/visda-2019/clipart/ --pretrained -b 1024 --evaluate --workers 25 >logs/clipart_eval.log- Supports ad hoc evaluation and BN adaptation.
- The datasets for the different domains must be downloaded and the VisDA-2019 folder needs to be specified.
- Then, VisDA-2019 classes are mapped to ImageNet classes and symlinks are created in the main directory for the mapped classes. The symlink folders now have 1000 sub-directories corresponding to the 1000 ImageNet classes with symlinks pointing to the corresponding VisDA-2019 images. We refer to these symlink folders as ImageNet-D because it is a subset of VisDA-2019, mapped to ImageNet classes.