This project implements a handwritten digit recognition algorithm based on deep learning. On the basis of the completion of the basic algorithm, ① this project adds a progress bar of training, ② calls gpu for training, ③ saves the latest training weight, ④ draws the change curve of loss and accuracy in the training process in real time, ⑤ Added Squeeze-and-Excitation attention, the code notes are detailed, the structure is simple, and if you have any problems, please leave a message.
本项目实现了一种基于深度学习的手写体数字识别算法。在基本算法完成的基础上，①本项目增加了训练进度条，②调用GPU进行训练，③保存最新的训练权重，④实时绘制训练过程中损失和准确度的变化曲线，⑤增加了SE注意力机制，代码注释详细，结构简单，如有问题请留言。
2022.9.30 更新（update）
这是对本项目的一次更新，本次更新内容是：在原来的算法基础上，增加了注意力机制，注意力机制使用的是Squeeze-and-Excitation的方法，本人已在本地做了相应的测试，由于选取的是手写数字识别的数据集，可能效果提升并没有很明显，因为已经高达98％，不过这是一次关于手动添加模块，并进行类定义与传入参数的操作，是对网络结构改动的积极探索，欢迎大家的疑问和讨论
This is an update to this project, this update content is: on the basis of the original algorithm, the attention mechanism is added, the attention mechanism uses the Squeeze-and-Excitation Attention, I have done the corresponding test locally, due to the selection of handwritten numeral recognition data set, the effect may not be very obvious, because it has been as high as 98%, but this is about manually adding modules. And the operation of class definition and incoming parameters is a positive exploration of the changes in the network structure. Questions and discussions are welcome.