Pytorch Implementation for Paper: Learning to Compare: Relation Network for Few-Shot Learning
download mini-imagenet and make it looks like:
mini-imagenet/
├── images
├── n0210891500001298.jpg
├── n0287152500001298.jpg
...
├── test.csv
├── val.csv
└── train.csv
LearningToCompare-Pytorch/
├── compare.py
├── MiniImagenet.py
├── Readme.md
├── repnet.py
├── train.py
└── utils.pypython train.pycurrent code support multi-gpus on single machine training, to disable it and train on single machine,
just set device_ids=[0] and downsize batch size according to your gpu memory capacity.
make sure ckpt directory exists, otherwise mkdir ckpt.
| Model | Fine Tune | 5-way Acc. | 20-way Acc | ||
|---|---|---|---|---|---|
| 1-shot | 5-shot | 1-shot | 5-shot | ||
| Matching Nets | N | 43.56% | 55.31% | 17.31% | 22.69% |
| Meta-LSTM | 43.44% | 60.60% | 16.70% | 26.06% | |
| MAML | Y | 48.7% | 63.11% | 16.49% | 19.29% |
| Meta-SGD | 50.49% | 64.03% | 17.56% | 28.92% | |
| TCML | 55.71% | 68.88% | - | - | |
| Learning to Compare | N | 57.02% | 71.07% | - | - |
| Ours, similarity ensemble | N | 55.2% | 68.8% | ||
| Ours, feature ensemble | N | 55.2% | 70.1% |