Handwritten Tibetan Character Recognition with PaddlePaddle

This code uses the PaddlePaddle framework for training, prediction, and plotting of a Tibetan MNIST handwritten digit recognition model. The entire process can be divided into the following steps: 1. **Dataset Loading**: First, the dataset is downloaded from the Kesci platform and decompressed, then the original images are converted into normalized grayscale images. 2. **Model Definition and Training**: - A simple CNN network structure is defined. - The optimizer, loss function, and accuracy calculation method are set. - Using Padd