• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International conference on neural information processing;Annual conference of Asia-Pacific Neural Network Society >IMDB-Attire: A Novel Dataset for Attire Detection and Localization
【24h】

IMDB-Attire: A Novel Dataset for Attire Detection and Localization

机译：IMDB-Attire：用于服装检测和定位的新型数据集

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

People's attire, or their way of dressing defines not only their social status and personality but also affects the way people meet and greet them. Attire detection has many useful applications such as clothing preferences in diverse regions of the world could be monitored and quantified. This information is very valuable for fashion designers. Real-time clothing recognition can be useful for security surveillance, where information about an individual's clothes can be used to identify crime suspects. Recently, deep learning algorithms have shown promise in the field of object detection and recognition. These algorithms are data hungry and are only as good as the data they are trained on. In this work, we have focused on three tasks to address this problem. We created a unique dataset of ～8000 images from IMDBb.com (movie rating website) to address the challenge of real-world application of the algorithm training for attire detection. The dataset contains pictures from movies, making the dataset a good source of images from the wild. We manually labelled 60 different classes of attire. Then we focused on multiclass classification and attire object detection using customized deep learning architectures including YOLO and SSD. We achieved a mean Average Precision (mAP) of 64.14% and an Average Precision (AP) of 91.14% for top 5 classes on YOLO. Available at https://github.com/saadyousuf45/.
机译：人们的着装或着装方式不仅定义了他们的社会地位和个性，而且还影响着人们见面和打招呼的方式。服装检测具有许多有用的应用程序，例如可以监控和量化世界各地的服装偏好。这些信息对于时装设计师来说非常有价值。实时服装识别对于安全监视很有用，在安全监视中，有关个人服装的信息可用于识别犯罪嫌疑人。最近，深度学习算法在对象检测和识别领域显示出了希望。这些算法需要大量数据，并且仅与训练有据的数据一样好。在这项工作中，我们专注于解决此问题的三个任务。我们从IMDBb.com（电影评级网站）创建了一个约8000张图像的唯一数据集，以应对服装检测算法训练在现实世界中的应用挑战。数据集包含电影中的图片，使数据集成为野外图像的良好来源。我们手动标记了60种不同的服装类别。然后，我们重点研究了使用定制的深度学习架构（包括YOLO和SSD）进行的多类分类和服装对象检测。我们在YOLO上排名前5位的班级平均平均精确度（mAP）为64.14％，平均精确度（AP）为91.14％。可在https://github.com/saadyousuf45/获得。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号