A Multi-Mode Modulator for Multi-Domain Few-Shot Classification (ICCV'21)

Official code implementation for A Multi-Mode Modulator for Multi-Domain Few-Shot Classification (ICCV 2021)

Requirements:

pip install tensorflow==2.4
pip install torch torchvision
pip install gin-config
pip install simclr
GPU with 16GB+ memory

Set Enviroment:

(1) Download&Process Meta-Dataset following: 
    https://github.com/google-research/meta-dataset#downloading-and-converting-datasets
    
(2) Download&Process 3 extra datasets (Mnist, Cifar10, Cifar100) following: 
    https://github.com/cambridge-mlg/cnaps --> Installation --> 3. Install additional test datasets (MNIST, CIFAR10, CIFAR100)
    
(3) Set the PROJECT_ROOT, META_DATASET_ROOT, and META_RECORDS_ROOT in datareader/path.py
    ulimit -n 50000

Training:

python run_triM.py --learning_rate 2e-3 --feature_adaptation MahSpecCoop -T 150000 --tasks_per_batch=16 

Testing:

python run_triM.py --learning_rate 2e-3 --feature_adaptation MahSpecCoop -T 150000 --tasks_per_batch=16 --test_model_path TEST_MODEL_CKPT_PATH --mode test --test_datasets=traffic_sign 

Bibtex

If you use this code or results for your research, please consider citing:

@INPROCEEDINGS{yanbin21triM,
  title     = {A Multi-Mode Modulator for Multi-Domain Few-Shot Classification},
  author    = {Liu, Yanbin and Lee, Juho and Zhu, Linchao and Chen, Ling and Shi, Humphrey and Yang, Yi},
  booktitle = {ICCV},
  year      = {2021}
}