cgarciae/point-cloud-mnist

# Point CLoud MNIST

A point cloud version of the original MNIST.


## Getting Started

```python
import matplotlib.pyplot as plt
import numpy as np

from datasets import load_dataset

# load dataset
dataset = load_dataset("cgarciae/point-cloud-mnist")
dataset.set_format("np")

# get numpy arrays
X_train = dataset["train"]["points"]
y_train = dataset["train"]["label"]
X_test = dataset["test"]["points"]
y_test = dataset["test"]["label"]

# plot some training samples
figure = plt.figure(figsize=(10, 10))
for i in range(3):
for j in range(3):
k = 3 * i + j
plt.subplot(3, 3, k + 1)
idx = np.random.randint(0, len(X_train))

plt.title(f"{y_train[idx]}")
plt.scatter(X_train[idx, :, 0], X_train[idx, :, 1])

plt.show()
```

## Format

* `points`: `(batch, point, 3)` array of uint8.
* `label`: `(batch, 1)` array of uint8.

Where `point` is the number of points in the point cloud. Points have no order and were shuffled when creating the data. Each point has the structure `[x, y, v]` where:

* `x`: is the x coordinate of the point in the image.
* `y`: is the y coordinate of the point in the image.
* `v`: is the value of the pixel at the point in the image.

Samples are padded with `0`s such that `point = 351` since its the largest number of non-zero pixels per image in the original dataset. You can tell apart padding point because they are the only ones where `v = 0`.

Here is the distribution of non-zero pixels in the MNIST:


原始信息汇总

数据集概述：Point Cloud MNIST

数据集描述

• 类型: 点云版本的MNIST数据集。
• 数据结构:
  • points: (batch, point, 3) 的 uint8 数组。
  • label: (batch, 1) 的 uint8 数组。

数据详情

• 点云结构: 每个点包含 [x, y, v]，其中：
  • x: 点的x坐标。
  • y: 点的y坐标。
  • v: 点在图像中的像素值。
• 点数: 每个样本的点数固定为351，通过填充 0 实现。
• 填充标识: 填充点可通过 v = 0 识别。

数据分布

• 非零像素分布: 展示了MNIST数据集中非零像素的分布情况。