Presentation | 2019-11-28 A proposal of a method for analyzing causes of incorrect detection when detecting objects using Deep Learning Tomonori Kubota, Takanori Nakao, Eiji Yoshida, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In this paper, we propose a method for analyzing the causes of incorrect detection / poor accuracy when detecting objects using Deep Learning. The authors have previously proposed a method for visualizing the cause of misrecognition in object recognition where one recognition object exists in an image. This time, we extended this method to object detection (YOLOv3) which predicts the existence position and classification probability of multiple objects. This method can extract and visualize causes of incorrect detection / poor accuracy, at pixel granularity. And, by applying the extracted information to the image in which the object cannot be correctly detected, it can be corrected to the image in which prediction of the position where the object exists and classification probability of correct class are improved. Thereby, the extracted information can correctly indicate the causes of incorrect detection / poor accuracy. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | object detection / convolutional neural network / inference / misdetection / visualizing / XAI |
Paper # | AI2019-30 |
Date of Issue | 2019-11-21 (AI) |
Conference Information | |
Committee | AI |
---|---|
Conference Date | 2019/11/28(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Naoki Fukuta(Shizuoka Univ.) |
Vice Chair | Yuichi Sei(Univ. of Electro-Comm.) / Yuko Sakurai(AIST) |
Secretary | Yuichi Sei(Osaka Univ.) / Yuko Sakurai(Tokyo Univ. of Agriculture and Technology) |
Assistant |
Paper Information | |
Registration To | Technical Committee on Artificial Intelligence and Knowledge-Based Processing |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A proposal of a method for analyzing causes of incorrect detection when detecting objects using Deep Learning |
Sub Title (in English) | |
Keyword(1) | object detection |
Keyword(2) | convolutional neural network |
Keyword(3) | inference |
Keyword(4) | misdetection |
Keyword(5) | visualizing |
Keyword(6) | XAI |
1st Author's Name | Tomonori Kubota |
1st Author's Affiliation | Fujitsu Laboratories LTD.(Fujitsu Lab.) |
2nd Author's Name | Takanori Nakao |
2nd Author's Affiliation | Fujitsu Laboratories LTD.(Fujitsu Lab.) |
3rd Author's Name | Eiji Yoshida |
3rd Author's Affiliation | Fujitsu Laboratories LTD.(Fujitsu Lab.) |
Date | 2019-11-28 |
Paper # | AI2019-30 |
Volume (vol) | vol.119 |
Number (no) | AI-317 |
Page | pp.pp.1-6(AI), |
#Pages | 6 |
Date of Issue | 2019-11-21 (AI) |