https://github.com/BoyuanJiang/matching-networks-pytorch

Matching Networks for one shot learning
https://github.com/BoyuanJiang/matching-networks-pytorch

one-shot-learning

Last synced: 13 days ago
JSON representation

Matching Networks for one shot learning

• Host: GitHub
• URL: https://github.com/BoyuanJiang/matching-networks-pytorch
• Owner: BoyuanJiang
• Created: 2017-10-25T14:37:15.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2017-10-26T07:32:05.000Z (about 7 years ago)
• Last Synced: 2024-07-31T23:44:42.800Z (3 months ago)
• Topics: one-shot-learning
• Language: Python
• Homepage:
• Size: 5.93 MB
• Stars: 228
• Watchers: 7
• Forks: 67
• Open Issues: 7
• Metadata Files:
  • Readme: README.MD

Awesome Lists containing this project

• awesome-few-shot-meta-learning - code (PyTorch)

README

        
This is the pytorch implement of **[Matching Networks for One Shot Learning](https://arxiv.org/pdf/1606.04080.pdf)**

![architecture](https://github.com/BoyuanJiang/matching-networks-pytorch/blob/master/architecture.png)

## Train 

If you want to train the model,simply run the code

```

python mainOmniglot.py 

```

You can set *fce = True* if you want use Full Context Embeddings.You can use the follow set in mainOmniglot.py for a 5-way one shot learning.

```

classes_per_set = 5

samples_per_class = 1

```

## Result

After about 30 epoches,you can achieve about 97% accuracy on train set and 96% on val and test set.

![result](https://github.com/BoyuanJiang/matching-networks-pytorch/blob/master/result.png)

## Acknowledgement

This work refer to [gitabcworld](https://github.com/gitabcworld/)'s implement,detail at[https://github.com/AntreasAntoniou/MatchingNetworks](https://github.com/AntreasAntoniou/MatchingNetworks).