Pedestrian attribute recognition
Pedestrian Attribute Recognition aims to develop a machine learning model to recognize person attributes including age, gender, clothing and accessories in a given person image. It is a challenging task due to high quality variance and inconsistent number of attributes in input image samples. Many m...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project (FYP) |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/139139 |
| _version_ | 1826112641856176128 |
|---|---|
| author | Liu, Yuan |
| author2 | Lin Zhiping |
| author_facet | Lin Zhiping Liu, Yuan |
| author_sort | Liu, Yuan |
| collection | NTU |
| description | Pedestrian Attribute Recognition aims to develop a machine learning model to recognize person attributes including age, gender, clothing and accessories in a given person image. It is a challenging task due to high quality variance and inconsistent number of attributes in input image samples. Many models have been proposed to tackle this problem. However, there are two open problems that remain to be addressed. On the one hand, visually similar pedestrian images tend to share similar attributes, and the upper and lower attributes normally match in style. However, the existing methods normally either ignore these relations or capture them ineffectively. On the other hand, how to maximise the benefit from each input pedestrian sample based on their importance to the machine learning model has not been explored, which can be critical given the huge variety in training images. In this report, we propose two deep learning based neural network models to tackle the two problems separately. The “Pentadent-Net” (PD-Net) is introduced with two parallel branches, where the two branches capture the relations among different samples and different attributes with a graphical method. The “Reinforced Sample Re-weighting” (RSR) is also proposed to re-weight samples in a batch during back-propagation through reinforcement learning, which can facilitate model training based on each input sample’s importance. The proposed PD-Net and RSR achieve state-of-the-art performance against existing methods on two large scale pedestrian attribute benchmarks PETA and PA-100K, which validates their effectives and advantages. The research paper generated from RSR has been accepted by WCCI 2020. The research paper on PD-Net is pending review for ICIP 2020. Notably, this project has been shortlisted as one of the 8 finalists for the EEE FYP Challenge 2020. |
| first_indexed | 2024-10-01T03:10:25Z |
| format | Final Year Project (FYP) |
| id | ntu-10356/139139 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T03:10:25Z |
| publishDate | 2020 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1391392023-07-07T18:52:03Z Pedestrian attribute recognition Liu, Yuan Lin Zhiping School of Electrical and Electronic Engineering ezplin@ntu.edu.sg Engineering::Electrical and electronic engineering::Computer hardware, software and systems Engineering::Computer science and engineering::Computer applications::Computer-aided engineering Pedestrian Attribute Recognition aims to develop a machine learning model to recognize person attributes including age, gender, clothing and accessories in a given person image. It is a challenging task due to high quality variance and inconsistent number of attributes in input image samples. Many models have been proposed to tackle this problem. However, there are two open problems that remain to be addressed. On the one hand, visually similar pedestrian images tend to share similar attributes, and the upper and lower attributes normally match in style. However, the existing methods normally either ignore these relations or capture them ineffectively. On the other hand, how to maximise the benefit from each input pedestrian sample based on their importance to the machine learning model has not been explored, which can be critical given the huge variety in training images. In this report, we propose two deep learning based neural network models to tackle the two problems separately. The “Pentadent-Net” (PD-Net) is introduced with two parallel branches, where the two branches capture the relations among different samples and different attributes with a graphical method. The “Reinforced Sample Re-weighting” (RSR) is also proposed to re-weight samples in a batch during back-propagation through reinforcement learning, which can facilitate model training based on each input sample’s importance. The proposed PD-Net and RSR achieve state-of-the-art performance against existing methods on two large scale pedestrian attribute benchmarks PETA and PA-100K, which validates their effectives and advantages. The research paper generated from RSR has been accepted by WCCI 2020. The research paper on PD-Net is pending review for ICIP 2020. Notably, this project has been shortlisted as one of the 8 finalists for the EEE FYP Challenge 2020. Bachelor of Engineering (Electrical and Electronic Engineering) 2020-05-16T05:17:43Z 2020-05-16T05:17:43Z 2020 Final Year Project (FYP) https://hdl.handle.net/10356/139139 en B3132-191 application/pdf Nanyang Technological University |
| spellingShingle | Engineering::Electrical and electronic engineering::Computer hardware, software and systems Engineering::Computer science and engineering::Computer applications::Computer-aided engineering Liu, Yuan Pedestrian attribute recognition |
| title | Pedestrian attribute recognition |
| title_full | Pedestrian attribute recognition |
| title_fullStr | Pedestrian attribute recognition |
| title_full_unstemmed | Pedestrian attribute recognition |
| title_short | Pedestrian attribute recognition |
| title_sort | pedestrian attribute recognition |
| topic | Engineering::Electrical and electronic engineering::Computer hardware, software and systems Engineering::Computer science and engineering::Computer applications::Computer-aided engineering |
| url | https://hdl.handle.net/10356/139139 |
| work_keys_str_mv | AT liuyuan pedestrianattributerecognition |