MSU Poultry Piling Dataset

The dataset consists of overhead images of brown hens in aviaries at Michigan State University. They are divided into positives (piling instances) and negatives (non-piling instances). They are further divided into train, val and test sets. It can be used to train an image-based piling detector. The number of images in each category are given in the table.

| Folder | Positives | Negatives | Total |
|--------|-----------|-----------|-------|
| Train | 2283 | 5279 | 7562 |
| Val | 261 | 592 | 853 |
| Test | 1276 | 1572 | 2848 |

Total number of images: 11263

## Piling Definition

A positive piling image is one containing a cluster of at least 10 chickens that are sufficiently tightly packed such that individual body outlines are not discernible. This is distinct from an image of crowded chickens that are not so tightly packed and their outlines are visible.

## Ethical Statement

Institutional Animal Care and Use Committee (IACUC) approval, PROTO202200159, was obtained at Michigan State University for experiments. All procedures were performed in compliance with Michigan laws and institutional guidelines.

本数据集包含密歇根州立大学（Michigan State University）禽舍内棕羽母鸡的俯视影像。数据集分为正样本（堆集实例）与负样本（非堆集实例），并进一步划分为训练集、验证集与测试集，可用于训练基于图像的堆集检测器。各分类下的图像数量见下表：
| 文件夹 | 正样本数 | 负样本数 | 总样本数 |
|--------|----------|----------|----------|
| 训练集 | 2283 | 5279 | 7562 |
| 验证集 | 261 | 592 | 853 |
| 测试集 | 1276 | 1572 | 2848 |

图像总数量：11263

### 堆集定义
正样本堆集影像指包含至少10只鸡的集群，且个体因堆叠过紧而轮廓无法辨识；此类影像与虽拥挤但未过度堆叠、个体轮廓仍可分辨的鸡群影像存在显著区别。

### 伦理声明
本实验已获得密歇根州立大学机构动物护理与使用委员会（Institutional Animal Care and Use Committee, IACUC）批准，批准编号为PROTO202200159。所有实验操作均符合密歇根州相关法律法规及机构指南要求。