This is an implementation for Zero-Shot Learning from Adversarial Feature Residual to Compact Visual Feature. The code will be released soon.
Data could be download here (g3ni).
Trained models could be download here (lzea) to evaluate the proposed method by running:
python evaluate_proto_feat.py --dataset APY --resSize 2048 --attSize 64 --nz 64 --syn_num 2000 --modeldir './APY_proto_model' --nepoch 40 --bs 128 --lr 0.0002
or:
python evaluate_proto_feat.py --dataset APY --resSize 1024 --attSize 64 --nz 64 --syn_num 2000 --modeldir './APY_proto1024_model' --nepoch 40 --bs 128 --lr 0.0002
Models could be trained by running:
python wgan_proto.py --dataset APY --resSize 2048 --attSize 64 --nz 64 --nepoch 5 --bs 512 --lr 0.00005 --modeldir './APY_proto_model' --syn_num 2000 --nepoch_cls 40 --bs_cls 128 --lr_cls 0.0002
or running code with feature selection:
python wgan_proto_sel.py --dataset APY --resSize 1024 --attSize 64 --nz 64 --nepoch 5 --bs 512 --lr 0.00005 --modeldir './APY_proto1024_model' --syn_num 2000 --nepoch_cls 40 --bs_cls 128 --lr_cls 0.0002
If you find this project useful for your research, please cite this paper:
@article{Liu2020ZeroShotLF, title={Zero-Shot Learning from Adversarial Feature Residual to Compact Visual Feature}, author={Bo Liu and Qiulei Dong and Zhanyi Hu}, journal={ArXiv}, year={2020}, volume={abs/2008.12962} }
If you have any questions, please contact [email protected]
This implementation refers to this repo and uses the datasets provided by http://datasets.d2.mpi-inf.mpg.de/xian/xlsa17.zip.