Publications

Publication

  <div class="row">
    <div class="col-sm-2 preview"><figure>

</figure> </div>

    <!-- Entry bib key -->
    <div id="thuan2024" class="col-sm-8">
    <!-- Title -->
    <div class="title">Point Central Transformer Network for Weakly-Supervised Point Cloud Semantic Segmentation</div>
    <!-- Author -->
    <div class="author">
    

    Anh-Thuan Tran</div>

    <!-- Journal/Book title and date -->
    
    
    <div class="periodical">
      <em>In Master’s Thesis</em>, 2024
    </div>
    <div class="periodical">
      
    </div>

      <!-- Links/Buttons -->
      <div class="links">
        <a class="abstract btn btn-sm z-depth-0" role="button">Abs</a>
      </div>
      
      <div class="badges">
        <span class="altmetric-embed" data-hide-no-mentions="true" data-hide-less-than="15" data-badge-type="2" data-badge-popover="right"
          
          ></span>
          <span class="__dimensions_badge_embed__"
          
            data-pmid=""
          
          data-hide-zero-citations="true" data-style="small_rectangle" data-legend="hover-right" style="margin-bottom: 3px;"></span>
      </div>

      <!-- Hidden abstract block -->
      <div class="abstract hidden">
        <p>In 3D scene understanding, segmentation tasks have a significant contribution in preprocessing and simplifying background, especially in real-world situations. Despite this crucial role in diverse practical applications, it causes high costs in labeling vast points, even in hundreds of millions per sample. To address this challenge, we propose an end-to-end transformer network designed for 3D weakly-supervised semantic segmentation. Different from prior methods, our network utilizes central-based attention, which processes purely on 3D points to resolve limited point annotations. Through integrating two embedding processes, the attention mechanism incorporates global features, thereby improving the representations of unlabeled points. In other words, our method establishes bidirectional connections between central points and their respective neighborhoods. To enrich geometric features and point position during training, we also implement position encoding. Point Central Transformer consistently reaches outstanding performance across various labeled point levels without requiring additional supervision. Experimental results on publicly available datasets such as S3DIS, ScanNet-V2, and STPLS3D underscore the superiority of our proposed approach compared to other state-of-the-art studies.</p>
      </div>
    </div>
  </div>

</li>

PointCT: Point Central Transformer Network for Weakly-Supervised Point Cloud Semantic Segmentation

Anh-Thuan Tran, Hoanh-Su Le, Suk-Hwan Lee, and 1 more author

In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2024

Abs HTML

Although point cloud segmentation has a principal role in 3D understanding, annotating fully large-scale scenes for this task can be costly and time-consuming. To resolve this issue, we propose Point Central Transformer (PointCT), a novel end-to-end trainable transformer network for weaklysupervised point cloud semantic segmentation. Divergent from prior approaches, our method addresses limited point annotation challenges exclusively based on 3D points through central-based attention. By employing two embedding processes, our attention mechanism integrates global features across neighborhoods, thereby effectively enhancing unlabeled point representations. Simultaneously, the interconnections between central points and their distinct neighborhoods are bidirectional cohered. Position encoding is further applied to enforce geometric features and improve overall performance. Notably, PointCT achieves outstanding performance under various labeled point settings without additional supervision. Extensive experiments on public datasets S3DIS, ScanNet-V2, and STPLS3D demonstrate the superiority of our proposed approach over other state-of-the-art methods.

</ol>

2023

1. 

   Local Graph Point Attention Network in Point Cloud Segmentation

   Anh-Thuan Tran, Hoanh-Su Le, Suk-Hwan Lee, and 1 more author

   IEEE Access, 2023

   Abs HTML

   Exploiting global factors and embedding them directly into local graphs in point clouds are challenging due to dense points and irregular structure. To accomplish this goal, we propose a novel end-to-end trainable graph attention network that extracts global features in terms of local graphs. Our network presents general local graph, which obtains the most fundamental features based on point order positions in different neighborhoods. Central point attention is introduced to share weights with neighboring points to reinforce central point impacts. As a result, one point in the specific local graph can obtain global features from corresponding ones in other neighborhoods but still focus on its area by encoded central point weight sharing. Experimental results on benchmark point cloud segmentation datasets demonstrate our proposed network’s competitive performance.

2. 

   General Local Graph Attention in Large-scale Point Cloud Segmentation

   Anh-Thuan Tran, Hoanh-Su Le, Oh-Joon Kwon, and 2 more authors

   In IEEE International Conference on Consumer Electronics, ICCE 2023, Las Vegas, NV, USA, January 6-8, 2023, 2023

   Abs HTML

   Due to the massive number of points and irregular structure, there is challenging to determine local graph relationships in large-scale point clouds. To address this problem, we propose a novel graph attention network that can embed global factors into neighboring points in specific local graphs. In this network, we introduce general local graph, which integrates all points sharing same order positions with encoded central point weights. In other words, it represents the most fundamental features of total local graphs. Besides, local graphs are regenerated to embed global factors by combining general local graph and central point attention. As a result, one point can obtain additional features from corresponding ones with same order positions in different neighborhoods. Experiment on large-scale point cloud segmentation dataset proves our network’s competitive performance.

3. KMMS

   Local Attention in Weakly Supervised Point Cloud Processing

   Anh-Thuan Tran, Hoanh-Su Le, Sang-Kyu Park, and 2 more authors

   In 2023 Korean Multimedia Society Spring Conference Proceedings Volume 26, No. 1, 2023

   Abs

   Point cloud segmentation has a vital position in multiple scenes understanding tasks on diverse practical applications, such as autonomous driving or robotics. However, labeling these vast points is highly time-consuming. To resolve this issue, we proposed an end-to-end trainable network focusing on limited points. Specifically, local attention is proposed to concentrate on relationships between query points and their neighboring ones to extract valuable features effectively. As a result, our network achieves competitive performance using only 1% number of labeled points and surpasses other studies, including methods with fully supervised settings.

2022

1. KMMS

   Local Graph Transformer in Semantic Point Cloud Segmentation

   Anh-Thuan Tran, Hoanh-Su Le, Suk-Hwan Lee, and 1 more author

   In 2022 Korean Multimedia Society Fall Conference Proceedings, Volume 25, No.2, 2022

   Abs

   The inner relationships between local graphs in large-scale point clouds is challenging to explore due to its irregular structure. To tackle this issue, we introduce graph transformer network that integrates global features into neighboring points. This network contains general local graph, which is an amalgation between different neighborhoods in the whole point cloud scene. In other words, it represents the most fundamental features in local graphs. Therefore, one point has two features: global factors from corresponding ones in different neighborhoods and its own geometric features.

2. KSEE

   Video Deepfake Detection using CNN and Vision Transformer

   Van-Nhan Tran, Eung-Joo Lee, Anh-Thuan Tran, and 1 more author

   In Korean Society of Electronic Engineers Conference, 2022

   Abs HTML

   Digital Transformation is one of the most crucial conversions in the 21st century. By using incredible technologies such as the Internet of Things, Artificial Intelligence, Cloud Computing, libraries nowadays can analyze and automatically answer student questions continuously. Understanding the opportunity, in this research, we propose a complete management system for smart library applied artificial intelligence (AI). Our contributions focus on the following main points. Firstly, our proposed system is a unified platform in which each AI model can interact and support the others. Secondly, we have a user-friendly dashboard that helps the managers simply operate this system. Finally, the platform is scalable because it can manage several libraries at the same time.

2021

1. 

   Design of an AI-based Smart Classroom Management System

   Tran Anh-Thuan

   In Bachelor’s Thesis, 2021

   Abs HTML

   Smart Classroom is one of the most potential solutions for smart education in the 21st century. By using incredible technologies such as the Internet of Things, Artificial Intelligence, Cloud Computing, classrooms nowadays can analyze and evaluate the learning performance continuously. Understanding this opportunity, we build a complete management system for smart classroom applied computer vision. Using the smart classroom solution, students can request for absence and be accepted or rejected online by their teachers. Besides, the dashboard also shows the summary of the present students and absent ones. Next, the managers or teachers can download the attendance list for further reports, the dashboard can filter by week, months, year, and each student. In addition, face recognition technology can use in security that automatically detects abnormal intrusions and notify the guards on time. Finally, the system can track the student’s emotions and visualize them on the chart in real-time. Based on this information, managers or teachers can evaluate their teaching performance. Our contributions focus on three following main points. Firstly, our proposed system can summarize data from several devices, analyze and visualize them on an interactive dashboard. In addition to the management system, the dashboard also has a user-friendly interface that can help managers and teachers who are unfamiliar with technologies can simply operate them. Finally, the AI Modules in the classroom are also utilized and improved. Specifically, the emotion detection model has an accuracy of 72.3% which is greater than the other applications.

2. 

   A proposed method for opinion mining online media of smartphone with Vietnamese text

   Anh-Thuan Tran, Dang-Huy Truong, Hoanh-Su Le, and 2 more authors

   In KMIS International Conference, 2021

   Abs HTML

   Digital Transformation is one of the most crucial conversions in the 21st century. By using incredible technologies such as the Internet of Things, Artificial Intelligence, Cloud Computing, libraries nowadays can analyze and automatically answer student questions continuously. Understanding the opportunity, in this research, we propose a complete management system for smart library applied artificial intelligence (AI). Our contributions focus on the following main points. Firstly, our proposed system is a unified platform in which each AI model can interact and support the others. Secondly, we have a user-friendly dashboard that helps the managers simply operate this system. Finally, the platform is scalable because it can manage several libraries at the same time.

2020

1. 

   A proposed method for opinion mining online media of smartphone with Vietnamese text

   Tran Anh-Thuan, Nguyen Anh-Nhat, Bui Xuan-Thanh, and 2 more authors

   In The 6th international Conference for Small and Medium Business 2020 (ICSMB 2020), Volume 1, No.7, 2020

   Abs HTML

   The rapid growth of the Internet has increased the number of people who accessed on Internet, customers can quickly search for information about products. Besides, businesses can clearly promote their products to users with low cost by many methods, but the most typical one is online newspapers and reviews. This study classify entire news (with user’s comments) in smartphone field. To accomplish that, four machine learning models are considered (LSTM, BiLSTM, SVM, and BERT). One of them is based on BERT that is a modern Deep Learning model for Natural Language Processing tasks created by Google recently that has achieved state-of-art results.

2019

1. 

   SmartHome IoT Application Development

   Tran Anh-Thuan, Truong Dang-Huy, Nguyen Hoang-Dung, and 2 more authors

   In The 15th International Conference on Multimedia Information Technology and Application Volume 1, 2019

   Abs HTML

   In recent years, Vietnamese living facility has risen along with their spiritual needs. People concern more about their spiritual life, including SmartHome. Using technology, having on-time warnings and saving energy, SmartHome brings convenience and safety to people. However, SmartHomes in Viet Nam are mostly provided by abroad business organizations. This scientific research focuses on building some basic features of SmartHome such as controlling devices or thief warnings with low cost, match with Viet Nam environment and culture.

–>