MAML-Pytorch

PyTorch implementation of the supervised learning experiments from the paper: Model-Agnostic Meta-Learning (MAML).

Version 1.0: Both MiniImagenet and Omniglot Datasets are supported! Have Fun~

Version 2.0: Re-write meta learner and basic learner. Solved some serious bugs in version 1.0.

For Tensorflow Implementation, please visit official HERE and simplier version HERE.

For First-Order Approximation Implementation, Reptile namely, please visit HERE.

Platform

• python: 3.x
• Pytorch: 0.4+

MiniImagenet

Howto

For 6-way 1-shot exp., it allocates nearly 1GB GPU memory.

1. download Fruits fresh and rotten for classification dataset from here
2. extract it like:

dataset/
├── train/
	freshapples/
		├── n0210891500001298.jpg  
		├── n0287152500001298.jpg 
		...
	freshbanana/
	rottenapples/
	rottenbanana/
├── test/
	freshoranges/
	freshoranges/

3. modify the arguments in terminal:

        usage: rotten_train.py [-h] [--n_way N_WAY] [--k_spt K_SPT] [--k_qry K_QRY]
                       [--path PATH] [--batch BATCH] [--epoch EPOCH]
                       [--imgsz IMGSZ] [--imgc IMGC] [--task_num TASK_NUM]
                       [--meta_lr META_LR] [--update_lr UPDATE_LR]
                       [--update_step UPDATE_STEP]
                       [--update_step_test UPDATE_STEP_TEST]

optional arguments:
  -h, --help            show this help message and exit
  --n_way N_WAY         n way
  --k_spt K_SPT         k shot for support set
  --k_qry K_QRY         k shot for query set
  --path PATH           path to the dataset
  --batch BATCH         epoch number
  --epoch EPOCH         epoch number
  --imgsz IMGSZ         imgsz
  --imgc IMGC           imgc
  --task_num TASK_NUM   meta batch size, namely task num
  --meta_lr META_LR     meta-level outer learning rate
  --update_lr UPDATE_LR
                        task-level inner update learning rate
  --update_step UPDATE_STEP
                        task-level inner update steps
  --update_step_test UPDATE_STEP_TEST
                        update steps for finetunning

to your actual data path.

4. just run python rotten_train.py and the running screenshot is as follows:

If your reproducation perf. is not so good, maybe you can enlarge your training epoch to get longer training. And MAML is notorious for its hard training. Therefore, this implementation only provide you a basic start point to begin your research. and the performance below is true and achieved on my machine.

References

@misc{MAML_Pytorch,
  author = {Liangqu Long},
  title = {MAML-Pytorch Implementation},
  year = {2018},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/dragen1860/MAML-Pytorch}},
  commit = {master}
}