Presentation | 2016-10-06 A Method for Predicting Image Region Specified by Query Texts Kou Endo, Kotarou Funakoshi, Eric Nichols, Masaki Aono, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We propose a method to identify the area in the image that corresponds to a user's text query. The input to the system is the image and the text representing the area. The output from the system is the bounding box (x, y, w, h), where (x, y) is the upper-left corner coordinates of the area, w is its width, and h is its height. We treat area prediction as a regression problem and train deep neural network model directly the input image and query text, eliminating the need for external candidate region prediction. In evaluation on the "ReferIt" dataset, provided by the University of North Carolina, our proposed approach achieves state-of-the-art performance, surpassing a baseline system that learns independent regressions model for each of the four parameters and the candidate generation and ranking approach of Hu et al.[2] |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | deep learning / query texts / bounding box / regression |
Paper # | IE2016-66 |
Date of Issue | 2016-09-29 (IE) |
Conference Information | |
Committee | IE / ITE-ME / ITE-AIT |
---|---|
Conference Date | 2016/10/6(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Seishi Takamura(NTT) / Miki Haseyama(Hokkaido Univ.) / Tokiichiro Takahashi(TDU) |
Vice Chair | Takayuki Hamamoto(Tokyo Univ. of Science) / Atsuro Ichigaya(NHK) / Norio Tagawa(Tokyo Metropolitan Univ.) |
Secretary | Takayuki Hamamoto(NTT) / Atsuro Ichigaya(Chiba Inst. of Tech.) / Norio Tagawa |
Assistant | Kei Kawamura(KDDI R&D Labs.) / Keita Takahashi(Nagoya Univ.) |
Paper Information | |
Registration To | Technical Committee on Image Engineering / Technical Group on Media Engineering / Technical Group on Artistic Image Technology |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Method for Predicting Image Region Specified by Query Texts |
Sub Title (in English) | |
Keyword(1) | deep learning |
Keyword(2) | query texts |
Keyword(3) | bounding box |
Keyword(4) | regression |
1st Author's Name | Kou Endo |
1st Author's Affiliation | Toyohashi University of Technology(TUT) |
2nd Author's Name | Kotarou Funakoshi |
2nd Author's Affiliation | Honda Research Institute Japan(HRI-J) |
3rd Author's Name | Eric Nichols |
3rd Author's Affiliation | Honda Research Institute Japan(HRI-J) |
4th Author's Name | Masaki Aono |
4th Author's Affiliation | Toyohashi University of Technology(TUT) |
Date | 2016-10-06 |
Paper # | IE2016-66 |
Volume (vol) | vol.116 |
Number (no) | IE-239 |
Page | pp.pp.7-12(IE), |
#Pages | 6 |
Date of Issue | 2016-09-29 (IE) |