Presentation | 2022-03-09 Improving Weakly Supervised Instance Segmentation by Encoding Motion Information via Optical Flow Jun Ikeda, Junichiro Mori, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Weakly supervised instance segmentation is an important task that can significantly reduce the annotation cost of model training. Previous works have been using appearance information obtained from a single image for detection and segmentation. However, it poses the challenge of identifying objects with non-discriminative appearance. In this paper, we tackle this problem by using motion information from image sequences. Specifically, we propose a two-stream encoder that appropriately leverages appearance and motion features extracted from images and optical flows. In addition, we employ a pairwise loss that considers both appearance and motion information to supervise segmentation. We conduct extensive evaluations on the YouTube-VIS 2019 dataset. Our results demonstrate that the proposed method improves the AP of the state-of-the-art method by 3.1. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | instance segmentation / weakly supervised learning / optical flow |
Paper # | IMQ2021-15,IE2021-77,MVE2021-44 |
Date of Issue | 2022-03-02 (IMQ, IE, MVE) |
Conference Information | |
Committee | CQ / IMQ / MVE / IE |
---|---|
Conference Date | 2022/3/9(3days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online (Zoom) |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Media of five senses, Multimedia, Media experience, Picture codinge, Image media quality, Network,quality and reliability, etc |
Chair | Jun Okamoto(NTT) / Kenya Uomori(Osaka Univ.) / Masayuki Ihara(RIKEN) / Kazuya Kodama(NII) |
Vice Chair | Takefumi Hiraguri(Nippon Inst. of Tech.) / Gou Hasegawa(Tohoku Univ.) / Mitsuru Maeda(Canon) / Kiyoshi Kiyokawa(NAIST) / Hiroyuki Bandoh(NTT) / Toshihiko Yamazaki(Univ. of Tokyo) |
Secretary | Takefumi Hiraguri(NTT) / Gou Hasegawa(Ritsumeikan Univ.) / Mitsuru Maeda(Nagoya Univ.) / Kiyoshi Kiyokawa(NTT) / Hiroyuki Bandoh(Oosaka Inst. of Tech.) / Toshihiko Yamazaki(NTT) |
Assistant | Yoshiaki Nishikawa(NEC) / Ryoichi Kataoka(KDDI Research) / Kimiko Kawashima(NTT) / Masato Tsukada(NEC) / Takashi Yamazoe(Seikei Univ.) / Naoya Isoyama(NAIST) / Takenori Hara(DNP) / Mitsuhiro Goto(NTT) / Shunsuke Iwamura(NHK) / Shinobu Kudo(NTT) |
Paper Information | |
Registration To | Technical Committee on Communication Quality / Technical Committee on Image Media Quality / Technical Committee on Media Experience and Virtual Environment / Technical Committee on Image Engineering |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Improving Weakly Supervised Instance Segmentation by Encoding Motion Information via Optical Flow |
Sub Title (in English) | |
Keyword(1) | instance segmentation |
Keyword(2) | weakly supervised learning |
Keyword(3) | optical flow |
1st Author's Name | Jun Ikeda |
1st Author's Affiliation | The University of Tokyo(UT) |
2nd Author's Name | Junichiro Mori |
2nd Author's Affiliation | The University of Tokyo(UT) |
Date | 2022-03-09 |
Paper # | IMQ2021-15,IE2021-77,MVE2021-44 |
Volume (vol) | vol.121 |
Number (no) | IMQ-420,IE-422,MVE-423 |
Page | pp.pp.27-32(IMQ), pp.27-32(IE), pp.27-32(MVE), |
#Pages | 6 |
Date of Issue | 2022-03-02 (IMQ, IE, MVE) |