Introduction

This code provides an initial version for the implementation of the Neurocomputing paper "Invariant and Consistent: Unsupervised Representation Learning for Few-Shot Visual Recognition''.

Paper Link.

Prerequisites

Python 3.6+ Pytorch 1.4+ CUDA 10.1

Training

python main.py -a convnet --lr 0.03 --batch-size 128
--dist-url 'tcp://localhost:10001'
--multiprocessing-distributed --world-size 1 --rank 0
/your dataset path/

Testing

python eval.py
-a convnet --n_shots 20 --n_queries 5 --n_test_runs 600
--pretrained /model path/
--dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0
/your dataset path/