https://scholars.lib.ntu.edu.tw/handle/123456789/413012
Title: | Filter-invariant image classification on social media photos | Authors: | Chen Y.-H. Chao T.-H. Bai S.-Y. Lin Y.-L. WEN-CHIN CHEN WINSTON HSU |
Keywords: | Convolutional Neural Network (CNN); Filter Bias; Image Classification; Photo Filter; Siamese Network | Issue Date: | 2015 | Start page/Pages: | 855-858 | Source: | MM 2015 - Proceedings of the 2015 ACM Multimedia Conference | Abstract: | With the popularity of social media nowadays, tons of photos are uploaded everyday. To understand the image content, image classification becomes a very essential technique for plenty of applications (e.g., object detection, image caption generation). Convolutional Neural Network (CNN) has been shown as the state-of-The-Art approach for image classi-fication. However, one of the characteristics in social media photos is that they are often applied with photo filters, especially on Instagram. We find that prior works do not aware of this trend in social media photos and fail on filtered images. Thus, we propose a novel CNN architecture that utilizes the power of pairwise constraint by combining Siamese network and the proposed adaptive margin contrastive loss with our discriminative pair sampling method to solve the problem of filter bias. To the best of our knowledge, this is the first work to tackle filter bias on CNN and achieve stateof-the-Art performance on a filtered subset of ILSVRC2012. ? 2015 ACM. |
URI: | https://scholars.lib.ntu.edu.tw/handle/123456789/413012 https://www.scopus.com/inward/record.uri?eid=2-s2.0-84962787080&doi=10.1145%2f2733373.2806348&partnerID=40&md5=b002a75ba3cba1511466ea393b9833c8 |
ISBN: | 9781450334594 | DOI: | 10.1145/2733373.2806348 | SDG/Keyword: | Bandpass filters; Convolution; Neural networks; Social networking (online); Convolutional neural network; Filter Bias; Filtered images; Pairwise constraints; Photo Filter; Sampling method; State-of-the-art approach; State-of-the-art performance; Image classification |
Appears in Collections: | 資訊工程學系 |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.