Class-Inherent Transformations

Paper: FuCiTNet: Improving the generalization of deep learning networks by the fusion of learned class-inherent transformations

Requirements

This code was tested using python 3.6.9, cuda 10 and PyTorch 1.1.0

Run the following line pip3 install -r ./requirements.txt to install the requirements.

Data

Data must present the following layout

dataset/
    train/
        class1/
            img1.jpg
            ...
        class2/
            imga.jpg
            ...
    test/
        class1/
            img3.jpg
            ...
        class2/
            imgc.jpg
            ...

Datasets used in the paper are available within the folder data

Training

In order to train the model: python3 src/train.py --all_lambdas --classifier_name resnet18 --dataset church_vs_palace --data_size 64

Testing

Trained models on "church vs palace", "cat vs dog" and "cat vs dog vs goldfish" are provided in weights (only the best lambdas reported in the paper):

• 0.05 for cat_vs_dog
• 0.005 for cat_vs_dog_vs_goldfish
• 0.0025 for church_vs_palace

To run the test script: python3 src/test.py --lambda_class 0.05 --weight_type checkpoint --classifier_name resnet18 --kfold 3 --dataset cat_vs_dog --data_size 64

Citation

If you use our code, please cite our work

@article{REYAREA2020188,
title = {{FuCiTNet}: Improving the generalization of deep learning networks by the fusion of learned class-inherent transformations},
journal = {Information Fusion},
volume = {63},
pages = {188 - 195},
year = {2020},
issn = {1566-2535},
doi = {https://doi.org/10.1016/j.inffus.2020.06.015},
url = {http://www.sciencedirect.com/science/article/pii/S1566253520303122},
author = {Manuel Rey-Area and Emilio Guirado and Siham Tabik and Javier Ruiz-Hidalgo}
}