spiking meta-learning

This repository contains the code for ''A Two-stage Spiking Meta-learning Method for Few-shot Classification''.

Main Results

The models on Omniglot use SNN-ConvNet-4 as backbone.

accuracy (%) on Omniglot

method	2-way 1-shot	2-way 5-shot	5-way 1-shot	5-way 5-shot	20-way 1-shot	20-way 5-shot
CESM	96.71±0.32	98.59±0.28	90.90 ±0.38	96.89±0.55	78.99±0.68	90.63±0.18
MESM	96.75±0.32	98.63±0.26	94.86±0.32	97.99±0.36	84.87±0.48	93.42±0.16

The models on miniImageNet and tieredImageNet use SNN-ResNet-12 as backbone, the channels in each block are 64-128-256-512, the backbone does NOT introduce any additional trick (e.g. DropBlock or wider channel in some recent work).

accuracy (%) on miniImageNet

method	2-way 1-shot	2-way 5-shot	5-way 1-shot	5-way 5-shot	10-way 1-shot	10-way 5-shot	20-way 1-shot	20-way 5-shot
CESM	75.13±0.30	84.65±0.21	48.37±0.24	65.61±0.26	34.26±0.26	52.60±0.65	23.14±0.42	38.94±0.43
MESM	74.56±0.24	84.68±0.19	51.54±0.23	69.94±0.18	35.87±0.84	53.83±0.79	25.03±0.47	41.08±0.44

accuracy (%) on tieredImageNet

method	2-way 1-shot	2-way 5-shot	5-way 1-shot	5-way 5-shot
CESM	75.98±0.21	86.46±0.26	52.51±0.22	68.66±0.28
MESM	76.59±0.18	88.49±0.13	53.76±0.15	69.01±0.16

Running the code

Preliminaries

Environment

• Python 3.7.3
• Pytorch 1.2.0
• tensorboardX

Datasets

• [Omniglot]
• miniImageNet (courtesy of Spyros Gidaris)
• tieredImageNet (courtesy of Kwonjoon Lee)

1. Training Classifier-Baseline

python train_cesm.py --config configs/train_classifier_mini.yaml

2. Training Meta-Baseline

python train_mesm.py --config configs/train_meta_mini.yaml

3. Test

To test the performance, modify configs/test_few_shot.yaml by setting load_encoder to the saving file of Classifier-Baseline, or setting load to the saving file of Meta-Baseline.

E.g., load: ./save/meta_mini-imagenet-1shot_meta-baseline-resnet12/max-va.pth

Then run

python test_few_shot.py --shot 1