EventMind

## 📥 EventMind Dataset Access

Thank you for your interest in the **EventMind** dataset.
To ensure responsible usage and maintain the fairness and integrity of the benchmark, users are required to provide basic information before gaining access.

### 📌 How to Apply for Access

Please send the following information:

- **Full Name**
- **Affiliation / Institution**
- **Laboratory or Research Group**
- **Intended Usage / Research Purpose**

Access will be granted after a brief review to ensure compliance with the dataset usage policy.

---

## 📦 Dataset Format

The EventMind dataset is stored in **NumPy `.npz` format**, where each file contains event stream data organized into temporal segments.

### 🔑 File Structure

Each `.npz` file contains a single key:

```python
import numpy as np

data = np.load('xxx.npz', allow_pickle=True)
print(data.files)
# ['event_bins']
```

- **`event_bins`**: an object array
- Each element corresponds to a **temporal window (event bin)**

> **Note:** The provided `event_bins` are preprocessed representations of the raw event stream. Specifically, the event data has been temporally aggregated into discrete time windows, enabling more efficient downstream processing and modeling.

---

### 🧩 Event Bin Structure

Each `event_bin` is a structured NumPy array with the following dtype:

```python
dtype = [
('p', 'u1'), # polarity
('t', '<i8'), # timestamp
('x', '<u2'), # x-coordinate
('y', '<u2') # y-coordinate
]
```

Each row represents a single **event**:

| Field | Type | Description |
|------|------|------------|
| `p` | uint8 | Polarity (0: negative, 1: positive) |
| `t` | int64 | Timestamp (in microseconds) |
| `x` | uint16 | Pixel x-coordinate |
| `y` | uint16 | Pixel y-coordinate |

---

### ⏱ Temporal Organization

- The full event stream is **partitioned into multiple event bins**
- Each bin corresponds to a **fixed time window** (e.g., 25 ms)

```python
event_bins = data['event_bins']
print(event_bins.shape)
# (N,)
```

- `N` = number of temporal segments
- Each segment contains a **variable number of events**

---

### 📊 Example

```python
event_bins = data['event_bins']

# Access the first temporal window
bin_0 = event_bins[0]

print(bin_0[:3])
```

---

### ⚠️ Notes

- The array uses `dtype=object`, so **`allow_pickle=True` is required**
- Each event bin contains a **variable number of events**

📥 EventMind 数据集获取权限

感谢您对**EventMind**数据集的关注。为确保负责任使用，维护基准测试的公平性与完整性，申请者需提供基础信息后方可获取数据集访问权限。

### 📌 申请获取权限的流程

请发送以下信息：
- **完整姓名**
- **所属单位/机构**
- **实验室或研究团队**
- **意向用途/研究目的**

经简短审核以确认符合数据集使用政策后，即可获得访问权限。

---

## 📦 数据集格式

EventMind 数据集以**NumPy `.npz` 格式**存储，每个文件包含按时间片段组织的事件流数据。

### 🔑 文件结构

每个 `.npz` 文件仅包含一个键：
python
import numpy as np

data = np.load('xxx.npz', allow_pickle=True)
print(data.files)
# ['event_bins']

- **`event_bins`**：一个对象数组
- 每个元素对应一个**时间窗口（event bin，事件分箱）**

> 注意：提供的 `event_bins` 是对原始事件流的预处理表征。具体而言，我们已将事件数据按时间聚合为离散的时间窗口，以提升下游处理与建模的效率。

---

### 🧩 事件分箱结构

每个 `event_bin` 是一个结构化 NumPy 数组，其数据类型（dtype）定义如下：
python
dtype = [
('p', 'u1'), # 极性（polarity）
('t', '<i8'), # 时间戳（timestamp）
('x', '<u2'), # X坐标（x-coordinate）
('y', '<u2') # Y坐标（y-coordinate）
]

每一行代表单个**事件**：
| 字段 | 类型 | 说明 |
|------|------|------|
| `p` | uint8 | 极性（0表示负极性，1表示正极性） |
| `t` | int64 | 时间戳，单位为微秒 |
| `x` | uint16 | 像素X坐标 |
| `y` | uint16 | 像素Y坐标 |

---

### ⏱ 时间组织方式

- 完整的事件流被**划分为多个事件分箱**
- 每个分箱对应一个**固定时长的时间窗口**（例如25毫秒）
python
event_bins = data['event_bins']
print(event_bins.shape)
# (N,)

- `N` 表示时间片段的总数
- 每个片段包含**可变数量的事件**

---

### 📊 使用示例
python
event_bins = data['event_bins']

# 访问第一个时间窗口
bin_0 = event_bins[0]

print(bin_0[:3])

---

### ⚠️ 注意事项

- 该数组使用 `dtype=object`，因此**加载时需设置 `allow_pickle=True`**
- 每个事件分箱包含**可变数量的事件**