AutoEncoder with Pytorch

This project is an implementation of auto-encoder with MNIST dataset and pytorch = 1.4.0

Learn tensorboard by the way

Instruction to Prepare MNIST dataset

bash ./script/prepare.sh

Instruction to Train Model

bash ./script/train.sh ${model_name}

${model_name} can be one of conv/linear/vae...

Instruction to use tensorboard visulize

tensorboard --logdir=runs --bind_all

Trainning Loss

Decode Performance

Model Architecture

have to be aware of these information:

The output shape of ConvTranspose2d can be computed by formula:

Reference

Hung-yi Lee MachineLearning Spring2020 Course (https://www.bilibili.com/video/BV1JE411g7XF?p=59)

https://github.com/L1aoXingyu/pytorch-beginner (Conv Model from here, But train code with this repositories have a bug in transform operation, detail in this pr page)