Mushui Liu · Fangtai Wu · Bozheng Li · Ziqian Lu · Yunlong Yu ✉ · Xi Li

Zhejiang University

Please install the torch 1.13.1 and torchvision 0.14.1

Please download the dataset and put it in the correct path

We have given our pretrain weight in miniImageNet, but you can also train your own by performing our pretrain script:

python pretrain_clip_adapter.py --backbone_class Res12 --lr 0.1 --query 15

For example, to train the 1-shot 5-way model with Res12 backbone on MiniImageNet:

python train_fsl.py --max_epoch 50 --model_class MultiSem_Bfusion_Adapter --backbone_class Res12 --dataset MiniImageNet --way 5 --eval_way 5 --shot 1 --eval_shot 1 --query 15 --eval_query 15 --balance 0.01 --temperature 64 --temperature2 64 --lr 0.00001 --lr_mul 30 --lr_scheduler step --step_size 10 --gamma 0.5 --init_weights ./saves/initialization/miniimagenet/max_acc_sim_mixloss_TextAdapter.pth --save_dir ./else_check --eval_interval 1 --use_euclidean  --gpu 3 --seed 3

If you find our work, this repository, or pretrained models useful, please consider giving a star ⭐ and citation.

@inproceedings{liu2025envisioning,
  title={Envisioning class entity reasoning by large language models for few-shot learning},
  author={Liu, Mushui and Wu, Fangtai and Li, Bozheng and Lu, Ziqian and Yu, Yunlong and Li, Xi},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={18},
  pages={18906--18914},
  year={2025}
}

If you have any questions, please create an issue on this repository or contact at lms@zju.edu.cn.