twuilliam / domainnet-helpers

Helpers to pre-process the DomainNet dataset for cross-domain image retrieval.

domainnet-helpers

Helper scripts for open cross-domain retrieval experiments with the DomainNet dataset [website].

Image annotations

Run parse.py to create pandas dataframes for all domains of the dataset. There will be 2 dataframes: one for the quickdraw domain, one for the five other domains. Each entry in the dataframe consist of the filename path, the label and the split.

parse.py also contains the list of the 188 categories of DomainNet that overlap with ImageNet.

In our paper we consider the quickdraw domain as sketches, and the sketch domain as pencil sketches.

Word vectors

Run w2v.py to get the word vector for all class names. Word vectors are stored in a dictionary in a .npz file.

It requires the gensim library: conda install -c anaconda gensim

Image pre-processing

Run resize.py. Images from all domains will be resized to 224x224.

Citation

If you find these scripts useful, please consider citing our paper:

@article{
    Thong2020OpenSearch,
    title={Open Cross-Domain Visual Search},
    author={Thong, William and Mettes, Pascal and Snoek, Cees G.M.},
    journal={CVIU},
    year={2020},
    url={https://arxiv.org/abs/1911.08621}
}