Super multi-class object recognition system using a large amount of image data

2014 Fiscal Year Final Research Report

• Project/Area Number: 24680017
• Research Category: Grant-in-Aid for Young Scientists (A)
• Allocation Type: Partial Multi-year Fund
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: The University of Tokyo
• Principal Investigator: HARADA Tatsuya 東京大学, 情報理工学(系)研究科, 教授 (60345113)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 画像認識 / コンピュータビジョン / 機械学習 / 人工知能

Outline of Final Research Achievements

The goal of our research is the construction of the super multi-class generic object recognition system by learning the relationship between a large amount of image and text data statistically. A method to continuously learn the classifiers from a huge amount of data without breakdown is crucial to realize this system. If there are many objects in one image, it is important to recognize where and what they are. A cost to construct high quality training dataset is so expensive that reducing the construction cost is also crucial. Moreover, a technique to find novel classes is a bottleneck for the continuously growing recognition system. In this research, we have tackled the above mentioned topics and produced some results.

Free Research Field

知能機械情報学