LED_Mnist_4000

# 散斑数据集(Mnist + Letter)
两个类别(数字+字母)，共8000张

## 1.数据采集方式
散射介质：单面毛玻璃

参数：物距(23cm), 像距(2cm), 光阑直径(2.5mm)

使用LCD液晶板进行光调制

## 2.数据集使用方法
下载 dataset_4000.zip 并解压

数据由train, test, eval文件夹构成

train(训练集)共4000个散斑对，test+eval(测试和验证集)共100张散斑对，train中不包含test+eval

图像格式均为.png

图像经过裁剪+归一化之后为(512x512)

使用train作为训练集，eval作为训练时验证集，test作为模型测试集即可

# Speckle Dataset (Mnist + Letter)
This dataset contains two categories (digits + letters), with a total of 8000 speckle image pairs.

## 1. Data Collection Method
Scattering medium: single-sided frosted glass.
Parameters: object distance (23 cm), image distance (2 cm), aperture diameter (2.5 mm).
Light modulation was performed using an LCD liquid crystal panel.

## 2. Dataset Usage Instructions
Download the dataset_4000.zip file and extract it.
The dataset is composed of three folders: train, test, and eval.
The training set (train) includes 4000 speckle image pairs, while the test and eval sets (test + eval) have a total of 100 speckle image pairs, and the training set does not contain samples from the test and eval sets.
All images are in .png format.
After cropping and normalization, the image size is adjusted to 512×512.
It is recommended to use the train folder as the training set, the eval folder as the validation set during model training, and the test folder as the final model test set.