Prototypical Network

A re-implementation of Prototypical Network.

With ConvNet-4 backbone on miniImageNet.

Results in origin project

1-shot: 49.1% (49.4% in the paper)

5-shot: 66.9% (68.2% in the paper)

train__bak.py is the origin file, and train.py is modified from train_bak.py.
train.py can get more than 50% 1-shot-5-way acc!!!!

Results in this project

1-shot: 50.17% (49.4% in the paper)（you can get the result less than 100 epoch!）

5-shot: 68.11% (68.2% in the paper)

Environment

• python 3.6
• pytorch 1.2.0

Instructions

1. Download the images: https://drive.google.com/open?id=0B3Irx3uQNoBMQ1FlNXJsZUdYWEE

2. Make a folder materials/images and put those images into it.

--gpu to specify device for program.

1-shot Train

python train.py

1-shot Test

python test.py

5-shot Train

python train.py --shot 5 --train-way 30 --save-path ./save/proto-5

5-shot Test

python test.py --load ./save/proto-5/max-acc.pth --shot 5

to see the acc log

python show-logresult.py --load ./save/proto-5/trlog

Acknowledgment

The following repos are added in this work.

cyvius96/prototypical-network-pytorch: https://github.com/cyvius96/prototypical-network-pytorch