DFT-based Transformation Invariant Pooling Layer for Visual Classification

Jongbin Ryu, Ming-Hsuan Yang, and Jongwoo Lim

We propose a DFT based pooling layer for convolutional neural networks. The proposed DFT magnitude pooling satisfies translation invariance and shape preserving properties. It pools DFT magnitude of last convolution feature map based on shift theorem. Convolutional neural networks with the proposed method improve the performance of various visual classification tasks. We validate the ability of transformation invariance by sufficient experiments of the paper.

Paper

[pdf] [supp]

Code / Model

[MatConvNet] [PyTorch]

Model Comming soon.

Citation

Jongbin Ryu, Ming-Hsuan Yang and Jongwoo Lim, "DFT-based Transformation Invariant Pooling Layer for Visual Classification", in European Conference on Computer Vision, 2018.

Bibtex

@inproceedings{dft_2018_eccv,

author    = {Ryu, Jongbin and Yang, Ming-Hsuan and Lim, Jongwoo}, 

title     = {DFT-based Transformation Invariant Pooling Layer for Visual Classification}, 

booktitle = {European Conference on Computer Vision},

year      = {2018}

}

©2018 CVLab in HYU