ZhengPeng7/MovieNet-PS

---
license: cc-by-nc-sa-4.0
task_categories:
- image-classification
size_categories:
- 10B<n<100B
---

> MovieNet-PS is a dataset for person search in movie data.

Check our [GitHub repo](https://github.com/ZhengPeng7/GLCNet) for details.
Thanks to [MovieNet](https://movienet.github.io/), which is the source of raw data.

## Terms of Use

By downloading the dataset, you agree to the following terms:

1. You will use the data only for non-commercial research and educational purposes.

## Overall Information

## Introduction

'./Image': Images collected from the MovieNet dataset.

'./annotation/Images.mat': 1 * 736,835 struct (736,835 images)
Each line describes the pedestrian information of an image,
including the image name (imname), the number, and the location
of pedestrians appearing (nAppear and box) in this image.
'./annotation/pool.mat': test images.

'./annotation/test/train_test/Train.mat': 5532 query persons for training.
'./annotation/test/train_test/TestG2000-TestG10000.mat': 2900 query persons with gallery size varying from 2000 to 10000 for testing.

*Note: The location of each person is stored as (xmin, ymin, width, height),
i.e. crop_im = I ( idlocate(2):idlocate(2)+idlocate(4), idlocate(1):idlocate(1)+idlocate(3) );

## Citation

```
@article{zheng2021glcnet,
title={Global-local context network for person search},
author={Zheng, Peng and Qin, Jie and Yan, Yichao and Liao, Shengcai and Ni, Bingbing and Cheng, Xiaogang and Shao, Ling},
journal={arXiv preprint arXiv:2112.02500},
volume={8},
year={2021}
}
```

许可证：知识共享署名-非商业性使用-相同方式共享4.0（CC BY-NC-SA 4.0）
任务类别：图像分类（image-classification）
样本规模：100亿 < 样本总量 < 1000亿

> MovieNet-PS是一款面向影视数据的行人检索（person search）数据集。
详情请查阅我们的[GitHub仓库](https://github.com/ZhengPeng7/GLCNet)。
感谢原始数据源提供商[MovieNet](https://movienet.github.io/)。

## 使用条款
下载本数据集即视为您同意以下使用约定：
1. 您仅可将本数据集用于非商业性科研与教育用途。

## 总体信息
### 简介
- `./Image`：从MovieNet数据集中采集的图像数据。
- `./annotation/Images.mat`：1×736,835的结构体数组，共对应736,835张图像。每一行描述单张图像的行人信息，包括图像名称（imname）、该图像中出现的行人总数（nAppear）以及行人的位置框（box）。
- `./annotation/pool.mat`：测试集图像集合。
- `./annotation/test/train_test/Train.mat`：用于训练的5532个查询行人样本。
- `./annotation/test/train_test/TestG2000-TestG10000.mat`：用于测试的2900个查询行人样本，其候选图库（gallery）规模从2000至10000不等。

*注意：每个人物的位置以(xmin, ymin, width, height)格式存储，对应的图像裁剪代码为：`crop_im = I( idlocate(2):idlocate(2)+idlocate(4), idlocate(1):idlocate(1)+idlocate(3) );`

## 引用

@article{zheng2021glcnet,
title={用于行人检索的全局-局部上下文网络},
author={Zheng, Peng and Qin, Jie and Yan, Yichao and Liao, Shengcai and Ni, Bingbing and Cheng, Xiaogang and Shao, Ling},
journal={arXiv预印本 arXiv:2112.02500},
volume={8},
year={2021}
}