A re-implementation of Prototypical Network.
With ConvNet-4 backbone on miniImageNet.
For deep backbones (ResNet), see Meta-Baseline.
1-shot: 49.1% (49.4% in the paper)
5-shot: 66.9% (68.2% in the paper)
- python 3
- pytorch 0.4.0
-
Download the images: https://drive.google.com/open?id=0B3Irx3uQNoBMQ1FlNXJsZUdYWEE
-
Make a folder
materials/images
and put those images into it.
--gpu
to specify device for program.
python train.py
python test.py
python train.py --shot 5 --train-way 20 --save-path ./save/proto-5
python test.py --load ./save/proto-5/max-acc.pth --shot 5