| Presentation | 2018-09-20 Image Patchwork Data Augmentation Ryo Takahashi, Takashi Matsubara, Kuniaki Uehara, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Deep convolutional neural networks (CNNs) have demonstrated remarkable results thanks to their numerous parameters. However, such numerous parameters against the variety of training samples often have a risk of overfitting. To solve this problem, data augmentation methods have been proposed so far. Data augmentation increases the variety of data by flipping, cropping, resizing and re-colorizing, and it leads deep CNNs to achieve higher performance by preventing overfitting. In this study, we propose a new data augmentation technique called random image cropping and patching (RICAP), which randomly crops four images and patches them to create a new training image. We evaluate RICAP with current state-of-the-art CNNs (e.g., shake-shake regularization model) by comparison with competitive data augmentation techniques such as cutout and mixup; RICAP achieves a new state-of-the-art test error of 2.23% on CIFAR-10. We also confirm that deep CNNs with RICAP achieve better results on CIFAR-100 and ImageNet than those with other techniques. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Data Augmentation / Image Classification / Convolutional Neural Network |
| Paper # | PRMU2018-43,IBISML2018-20 |
| Date of Issue | 2018-09-13 (PRMU, IBISML) |
| Conference Information | |
| Committee | PRMU / IBISML / IPSJ-CVIM |
|---|---|
| Conference Date | 2018/9/20(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Shinichi Sato(NII) / Hisashi Kashima(Kyoto Univ.) |
| Vice Chair | Yoshihisa Ijiri(Omron) / Toru Tamaki(Hiroshima Univ.) / Masashi Sugiyama(Univ. of Tokyo) / Koji Tsuda(Univ. of Tokyo) |
| Secretary | Yoshihisa Ijiri(NEC) / Toru Tamaki(Osaka Univ.) / Masashi Sugiyama(Nagoya Inst. of Tech.) / Koji Tsuda(AIST) |
| Assistant | Go Irie(NTT) / Yoshitaka Ushiku(Univ. of Tokyo) / Tomoharu Iwata(NTT) / Shigeyuki Oba(Kyoto Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Pattern Recognition and Media Understanding / Technical Committee on Infomation-Based Induction Sciences and Machine Learning / Special Interest Group on Computer Vision and Image Media |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Image Patchwork Data Augmentation |
| Sub Title (in English) | |
| Keyword(1) | Data Augmentation |
| Keyword(2) | Image Classification |
| Keyword(3) | Convolutional Neural Network |
| 1st Author's Name | Ryo Takahashi |
| 1st Author's Affiliation | Kobe University(Kobe Univ.) |
| 2nd Author's Name | Takashi Matsubara |
| 2nd Author's Affiliation | Kobe University(Kobe Univ.) |
| 3rd Author's Name | Kuniaki Uehara |
| 3rd Author's Affiliation | Kobe University(Kobe Univ.) |
| Date | 2018-09-20 |
| Paper # | PRMU2018-43,IBISML2018-20 |
| Volume (vol) | vol.118 |
| Number (no) | PRMU-219,IBISML-220 |
| Page | pp.pp.47-54(PRMU), pp.47-54(IBISML), |
| #Pages | 8 |
| Date of Issue | 2018-09-13 (PRMU, IBISML) |