Pytorch code for following paper:
- Title : Transductive Few-shot Learning with Meta-Learned Confidence.
- Author : Seong Min Kye, Hae Beom Lee, Hoirin Kim, Sung Ju Hwang
- Python 3.6
- Pytorch 1.3.1
- miniImageNet 5-way 1-shot/5-shot
# miniImageNet, 5-way 1-shot
$ python train.py --is_train True --gpu 0 --transductive False --flip False --drop False --n_shot 1 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive False --flip False --drop False --n_shot 1 --n_train_class 15
# miniImageNet, 5-way 5-shot
$ python train.py --is_train True --gpu 0 --transductive False --flip False --drop False --n_shot 5 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive False --flip False --drop False --n_shot 5 --n_train_class 15
2 tieredImageNet 5-way 1-shot/5-shot
# tieredImageNet, 5-way 1-shot
$ python train.py --is_train True --gpu 0 --transductive False --flip False --drop False --n_shot 1 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive False --flip False --drop False --n_shot 1 --n_train_class 15
# tieredImageNet, 5-way 5-shot
$ python train.py --is_train True --gpu 0 --transductive False --flip False --drop False --n_shot 5 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive False --flip False --drop False --n_shot 5 --n_train_class 15
- miniImageNet 5-way 1-shot/5-shot
# miniImageNet, 5-way 1-shot
$ python train.py --is_train True --gpu 0 --transductive True --flip True --drop True --n_shot 1 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive True --flip True --drop True --n_shot 1 --n_train_class 15
# miniImageNet, 5-way 5-shot
$ python train.py --is_train True --gpu 0 --transductive True --flip True --drop True --n_shot 5 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive True --flip True --drop True --n_shot 5 --n_train_class 15
- tieredImageNet 5-way 1-shot/5-shot
# tieredImageNet, 5-way 1-shot
$ python train.py --is_train True --gpu 0 --transductive True --flip True --drop True --n_shot 1 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive True --flip True --drop True --n_shot 1 --n_train_class 15
# tieredImageNet, 5-way 5-shot
$ python train.py --is_train True --gpu 0 --transductive True --flip True --drop True --n_shot 5 --n_train_class 15
$ python train.py --is_train False --gpu 0 --transductive True --flip True --drop True --n_shot 5 --n_train_class 15
This code is based on the implementation of TapNet, MetaOptNet. And we use the dataset from MetaOptNet.