Presentation | 2019-03-01 Data Augmentation based on RGB-D SLAM 3D Data Ken Miyamoto, Osamu Tsukahara, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Large data preparation is inevitable for training a deep neural network. Preparing the large data with handcrafted label is not practical for recognizing an arbitrary object. Hence, various data augmentation methods have been published in the recent years. One of the representatives uses a computer graphics model. It can create images rendered from viewpoints under various light conditions. The second one augments images by translation, rotation or scaling. It is indispensable to train images taken from various viewpoints for recognizing an object on a natural image taken from an arbitrary viewpoint. However, the conventional data augmentations can’t create the images. We propose data augmentation based on 3D data taken through RGB-D SLAM. Our augmentation can create the images with low human cost. A user directs a recognition target area from an image obtained through RGB-D SLAM. The directed area is transformed to 3D positions using pose and depth. Additionally, the 3D positions are projected to all other images for obtaining corresponding directed areas. When training images obtained by the direction, the result shows that the mIoU is better than training images augmented by mix of translation, rotation and flipping. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Data Augmentation / Simultaneous Localization And Mapping / SLAM |
Paper # | PRMU2018-124,CNR2018-47 |
Date of Issue | 2019-02-21 (PRMU, CNR) |
Conference Information | |
Committee | PRMU / CNR |
---|---|
Conference Date | 2019/2/28(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Shinichi Sato(NII) / Tetsuo Ono(Hokkaido Univ.) |
Vice Chair | Yoshihisa Ijiri(Omron) / Toru Tamaki(Hiroshima Univ.) / Masayuki Kanbara(NAIST) / Kazunori Takashio(Keio Univ.) |
Secretary | Yoshihisa Ijiri(NEC) / Toru Tamaki(Osaka Univ.) / Masayuki Kanbara(Hokkaido Univ.) / Kazunori Takashio(Panasonic) |
Assistant | Go Irie(NTT) / Yoshitaka Ushiku(Univ. of Tokyo) / Wataru Mito(SECOM) / Yuka Kobayashi(Toshiba) / Tatsuya Ishihara(NTT) |
Paper Information | |
Registration To | Technical Committee on Pattern Recognition and Media Understanding / Technical Committee on Cloud Network Robotics |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Data Augmentation based on RGB-D SLAM 3D Data |
Sub Title (in English) | |
Keyword(1) | Data Augmentation |
Keyword(2) | Simultaneous Localization And Mapping |
Keyword(3) | SLAM |
1st Author's Name | Ken Miyamoto |
1st Author's Affiliation | Mitsubishi Electric Corporation(MELCO) |
2nd Author's Name | Osamu Tsukahara |
2nd Author's Affiliation | Mitsubishi Electric Corporation(MELCO) |
Date | 2019-03-01 |
Paper # | PRMU2018-124,CNR2018-47 |
Volume (vol) | vol.118 |
Number (no) | PRMU-459,CNR-460 |
Page | pp.pp.53-57(PRMU), pp.53-57(CNR), |
#Pages | 5 |
Date of Issue | 2019-02-21 (PRMU, CNR) |