2016 Fiscal Year Research-status Report
Fast, effective and robust person re-identification for large-scale real applications
| Project/Area Number |
15K16024
|
|---|
| Research Institution | Nara Institute of Science and Technology |
|---|
Principal Investigator |
伍 洋 奈良先端科学技術大学院大学, 研究推進機構, 特任助教 (30750559)
|
|---|
| Project Period (FY) |
2015-04-01 – 2018-03-31
|
| Keywords | Person re-identification / Metric learning / Deep learning / Dictionary learning / Scalable |
|---|
| Outline of Annual Research Achievements |
In this year, we have published 2 journal papers and 1 conference paper, and there are 1 journal paper and 1 conference paper currently under review.
In the paper of "Joint Hierarchical Category Structure Learning and Large Scale Image Classification", we built a novel tree structure model together with deep features for scalable and effective image classification. It is a general model which can be applied to person re-identification too. In another journal paper entitled "Re-identification by Neighborhood Structure Metric Learning" we explored a new metric learning model for person re-identification. We showed an enhanced and more effective learning model in our conference paper.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
We have made significant progress in the 2nd year on exploring effective and scalable recognition models and we were able to publish 2 journal papers and 1 conference paper, with another 1 journal paper and 1 conference paper submitted for reviewing.
Besides the originally planned model, we were able to explore more promising new models (using end-to-end deep learning), which are presented in our papers under reviewing.
|
| Strategy for Future Research Activity |
Since we have already got much progress on exploring the effectiveness and scalability, we will focus more on the efficiency in the last year, and try to integrate things together. We will also do the planned experiments on transfer learning for testing the generalization ability of our models.
|
| Causes of Carryover |
The planned hardware purchasing plan has been mainly covered by other budgets of our lab. In the 2nd year we didn't have much travelling cost and publication cost,but we have more ongoing work to be published in the last year. So the budget is saved for the last year.
|
| Expenditure Plan for Carryover Budget |
We will have more students working on extending the work and we will publish more papers. So the budget will be used for supporting these students, as well as publishing and presenting our research outcomes.
|
