BSNet

Code release for the paper BSNet: Bi-Similarity Network for Few-shot Fine-grained Image Classification. (TIP2020)

Requirements

• python=3.6
• PyTorch=1.2+
• torchvision=0.4.2
• pillow=6.2.1
• numpy=1.18.1
• h5py=1.10.2

Dataset

CUB-200-2011

• Change directory to ./filelists/CUB
• run source ./download_CUB.sh

Train

• method: relationnet|CosineBatch|OurNet.
• n_shot: number of labeled data in each class （1|5）.
• train_aug: perform data augmentation or not during training.
• gpu: gpu id.

python ./train.py --dataset CUB  --model Conv4 --method relationnet --n_shot 5 --train_aug --gpu 0
python ./train.py --dataset CUB  --model Conv4 --method CosineBatch --n_shot 5 --train_aug --gpu 0
python ./train.py --dataset CUB  --model Conv4 --method OurNet      --n_shot 5 --train_aug --gpu 0

Save features

python ./save_features.py --dataset CUB  --model Conv4 --method relationnet --n_shot 5 --train_aug --gpu 0
python ./save_features.py --dataset CUB  --model Conv4 --method CosineBatch --n_shot 5 --train_aug --gpu 0
python ./save_features.py --dataset CUB  --model Conv4 --method OurNet      --n_shot 5 --train_aug --gpu 0

Test

python ./test.py --dataset CUB  --model Conv4 --method relationnet --n_shot 5 --train_aug --gpu 0
python ./test.py --dataset CUB  --model Conv4 --method CosineBatch --n_shot 5 --train_aug --gpu 0
python ./test.py --dataset CUB  --model Conv4 --method OurNet      --n_shot 5 --train_aug --gpu 0

results

<table> <tr> <td colspan="3" align='center'>CUB-200-2011</td> </tr> <tr> <td align='center'></td> <td align='center'>5-way 5-shot Accuracy (%)</td> <td align='center'>5-way 1-shot Accuracy (%)</td> </tr> <tr> <td align='center'>Relation Network</td> <td align='center'>77.87 &plusmn; 0.64</td> <td align='center'>63.94 &plusmn; 0.92</td> </tr> <tr> <td align='center'>Cosine Network</td> <td align='center'>77.86 &plusmn; 0.68</td> <td align='center'>65.04 &plusmn; 0.97</td> </tr> <tr> <td align='center'>BSNet (R&C)</td> <td align='center'><b>80.99 &plusmn; 0.63</b></td> <td align='center'><b>65.89 &plusmn; 1.00</b></td> </tr> </table>

Citation

If you find this paper useful in your research, please consider citing:

@ARTICLE{9293172,
  author={X. {Li} and J. {Wu} and Z. {Sun} and Z. {Ma} and J. {Cao} and J. -H. {Xue}},
  journal={IEEE Transactions on Image Processing}, 
  title={BSNet: Bi-Similarity Network for Few-shot Fine-grained Image Classification}, 
  year={2021},
  volume={30},
  number={},
  pages={1318-1331},
  doi={10.1109/TIP.2020.3043128}}

References

Our code is based on Chen's contribution. Specifically, except for our core design, cosine network and BSNet, everything else （e.g. backbone, dataset, relation network, evaluation standards, hyper-parameters）are built on and integrated in https://github.com/wyharveychen/CloserLookFewShot.

Contact

Thanks for your attention! If you have any suggestion or question, you can leave a message here or contact us directly:

• spraisewu@gmail.com
• mazhanyu@bupt.edu.cn
• xiaoxulilut@gmail.com