Gestalt-Guided Image Understanding for Few-Shot Learning

This repo is based on LightningFSL.

To understand the code correctly, it is highly recommended to first quickly go through the pytorch-lightning documentation, especially LightningCLI. It won't be a long journey since pytorch-lightning is built on the top of pytorch.

✨ News

[Sep 17, 2022]

• GGIU is accepted to ACCV2022.

Installation

Just run the command:

git clone [email protected]:skingorz/GGIU.git
cd GGIU
conda env create -f env.yaml
conda activate baseCode

running an implemented few-shot model

1. Downloading Datasets:

   • miniImageNet

   The data format is as shown in dataset_and_process/datasets/miniImageNet.py:

2. Training:

   • Choose the corresponding configuration file in config (e.g.set_config_PN_train.py for PN model), set inside the parameter dataset directory, logging dir as well as other parameters you would like to change.
   • Change CONFIG_PY in train.sh (e.g., CONFIG_PY=config/set_config_PN_train.py).
   • To begin the running, run the command bash train.sh

3. Testing:

   • Choose the corresponding configuration file in config (e.g. set_config_PN_test.py for testing with GGIU), set inside the parameter dataset directory, logging dir, as well as other parameters you would like to change. If add GGIU, choose the configuration named with GGIU, set is_TTA to True, and set the value of lambd. Otherwise, set is_TTA to False.
   • Change CONFIG_PY and model in test.sh (e.g., CONFIG_PY=config/set_config_PN_TTA_test.py model=epoch=55-step=13999.ckpt).
   • To begin the testing, run the command bash test.sh