Testing codes for "One-Shot Image Classification by Learning to Restore Prototypes"
The repository contains the essential codes for RestoreNet and hope it will help to understand the idea.
-
You can do your own training, or alternatively, you may like to download the network parameters and pre-computed features at https://www.dropbox.com/sh/6jy0g8nfc97bvrm/AACORpPowNVnFXdwek7vUYjIa?dl=0
In the link, there are three files: 1) FeatureExtractor.pth 2) Transformer.pth 3) GalleryPool.
The first two .pth files are network parameters. Please put them in the folder ./Network Params/
The GalleryPool file is pre-computed images features used for self-training. We use it to avoid unnecessary computation. Please put it in ./
-
Put all images in miniImageNet dataset in the folder /miniImageNet/images.
-
Do evaluation by run 'python run_test.py'