Repository files navigation

Code for the NIPS 2017 paperPrototypical Networks for Few-shot Learning.

If you use this code, please cite our paper:

@inproceedings{snell2017prototypical,  title={Prototypical Networks for Few-shot Learning},  author={Snell, Jake and Swersky, Kevin and Zemel, Richard},  booktitle={Advances in Neural Information Processing Systems},  year={2017} }

• This code has been tested on Ubuntu 16.04 with Python 3.6 and PyTorch 0.4.
• InstallPyTorch and torchvision.
• Installtorchnet by runningpip install git+https://github.com/pytorch/tnt.git@master.
• Install the protonets package by runningpython setup.py install orpython setup.py develop.

• Runsh download_omniglot.sh.

• Runpython scripts/train/few_shot/run_train.py. This will run training and place the results intoresults.
  • You can specify a different output directory by passing in the option--log.exp_dir EXP_DIR, whereEXP_DIR is your desired output directory.
  • If you are running on a GPU you can pass in the option--data.cuda.
• Re-run in trainval modepython scripts/train/few_shot/run_trainval.py. This will save your model intoresults/trainval by default.

• Run evaluation as:python scripts/predict/few_shot/run_eval.py --model.model_path results/trainval/best_model.pt.