Readme.md

Transductive Few-Shot Classification on the Oblique Manifold

Introduction

The folders contain the code for paper Transductive Few-Shot Classification on the Oblique Manifold.

Environment

numpy==1.18.5
torch==1.7.0
Pillow==7.2.0
torchvision==0.8.0
tqdm==4.46.0

Train

For example:
5-way 5-shot with resnet18 in mini-ImageNet:
python train.py --n_way 5 --k_shot 5 --k_query 15 --skip False --dataset mini --backbone resnet18 --gpu 0,1,2\

5-way 1-shot with WRN in tiered-ImageNet:
python train.py --n_way 5 --k_shot 1 --k_query 15 --skip False --dataset tiered --backbone wideres --gpu 0,1,2\

Use the Pretrained Models

Move the models to folder checkpoint, and change the argument skip with value True:
For example:
5-way 5-shot with resnet18 in mini-ImageNet:
python train.py --n_way 5 --k_shot 5 --k_query 15 --skip True --dataset mini --backbone resnet18 --gpu 0,1,2\

5-way 1-shot with WRN in tiered-ImageNet:
python train.py --n_way 5 --k_shot 1 --k_query 15 --skip True --dataset tiered --backbone wideres --gpu 0,1,2\

Reference

• TIM
• Geoopt