This project is our implementation of Semantic-Aware Knowledge prEservation (SAKE) for zero-shot sketch-based image retrieval. More detailed descriptions and experimental results could be found in the paper.

If you find this project helpful, please consider citing our paper.

@inproceedings{liu2019semantic,
  title={Semantic-aware knowledge preservation for zero-shot sketch-based image retrieval},
  author={Liu, Qing and Xie, Lingxi and Wang, Huiyu and Yuille, Alan L},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={3662--3671},
  year={2019}
}

Download the resized TUBerlin Ext and Sketchy Ext dataset and our zeroshot train/test split files from here. Put the unzipped folder to the same directory of this project.

CSE-ResNet50 model with 64-d features on TUBerlin Ext:

python train_cse_resnet_tuberlin_ext.py

CSE-ResNet50 model with 64-d features on Sketchy Ext:

python train_cse_resnet_sketchy_ext.py

CSE-ResNet50 model with 64-d features on TUBerlin Ext:

python test_cse_resnet_tuberlin_zeroshot.py

CSE-ResNet50 model with 64-d features on Sketchy Ext:

python test_cse_resnet_sketchy_zeroshot.py

Our trained models and extracted zeroshot testing features can be downloaded from here.