NKUP, NKUP+
收藏资源简介:
NKUP数据集是一个用于行人再识别的新基准数据集,包含来自南开大学开放环境中的15个摄像头(8个室内和7个室外)的视频,通过行人检测器从监控视频中裁剪行人图像,获得了超过200,000个带有ID的边界框。数据集包含5,336个训练图像,332个查询图像和4,070个图库图像,每个行人平均有2.2种衣服。NKUP+数据集是一个长期且公开可用的行人再识别数据集,包含来自南开大学29个室内和室外摄像头的超过40,000个行人图像,每个行人提供多种图像,测试集根据不同外观分为几个子集。
The NKUP dataset serves as a novel benchmark for pedestrian re-identification, comprising video footage from 15 cameras (8 indoor and 7 outdoor) in an open environment at Nankai University. Pedestrian images were cropped from surveillance videos using a pedestrian detector, yielding over 200,000 ID-labeled bounding boxes. The dataset includes 5,336 training images, 332 query images, and 4,070 gallery images, with an average of 2.2 clothing variations per pedestrian. The NKUP+ dataset is a long-term, publicly available pedestrian re-identification dataset, featuring over 40,000 pedestrian images from 29 indoor and outdoor cameras at Nankai University. Each pedestrian is represented by multiple images, and the test set is divided into several subsets based on varying appearances.
NKUP Dataset Description
- Purpose: To facilitate research on person re-identification (re-ID) with diverse clothing changes.
- Collection Environment: 15 cameras (8 indoor and 7 outdoor) at Nankai University.
- Data Details:
- Total Bounding Boxes: Over 200,000.
- Training Set: 5,336 images of 40 identities.
- Query Set: 332 images.
- Gallery Set: 4,070 images of 67 identities.
- Average Clothing Variation: Each person has 2.2 kinds of clothes, excluding distractors.
NKUP+ Dataset Description
- Purpose: To address long-term person re-ID with cross-appearance changes.
- Collection Environment: 29 cameras (indoor and outdoor) at Nankai University.
- Data Details:
- Total Images: More than 40,000 pedestrian images.
- Test Set Structure: Divided into subsets based on appearance changes:
- Same-appearance gallery.
- Moderate cross-appearance gallery.
- Dramatic cross-appearance gallery (images with a time interval of more than three months).
Dataset Contents
- NKUP Dataset:
- Training Set: 5,336 images.
- Testing Set: 4,070 images.
- Query Set: 332 images.
- NKUP+ Dataset:
- Training Set: 22,590 images.
- Testing Set: Includes query, same appearance gallery, moderate appearance gallery, and dramatic appearance gallery.
Licensing
- Usage: For academic purposes only, no commercial use allowed.
- Copyright: © College of Computer Science, Nankai University.




