Instagram Filter Removal on Fashionable Images

TL;DR

This paper introduces IFRNet, a neural network designed to remove Instagram filters from images, improving social media image analysis by reversing style transfer effects and outperforming existing methods.

Contribution

The paper presents IFRNet, a novel approach that adaptively normalizes style information to effectively remove filters, enhancing the interpretability of filtered social media images.

Findings

IFRNet outperforms existing filter removal methods in accuracy.

The model effectively removes visual filter effects from images.

It also provides accurate filter classification and color estimation.

Abstract

Social media images are generally transformed by filtering to obtain aesthetically more pleasing appearances. However, CNNs generally fail to interpret both the image and its filtered version as the same in the visual analysis of social media images. We introduce Instagram Filter Removal Network (IFRNet) to mitigate the effects of image filters for social media analysis applications. To achieve this, we assume any filter applied to an image substantially injects a piece of additional style information to it, and we consider this problem as a reverse style transfer problem. The visual effects of filtering can be directly removed by adaptively normalizing external style information in each level of the encoder. Experiments demonstrate that IFRNet outperforms all compared methods in quantitative and qualitative comparisons, and has the ability to remove the visual effects to a great extent. Additionally, we present the filter classification performance of our proposed model, and analyze the dominant color estimation on the images unfiltered by all compared methods.