Go to file

snsd0805 687a994bea docs: update readme		2023-03-14 23:33:54 +08:00
images	docs: add readme	2023-03-14 23:28:08 +08:00
.gitignore	docs: update .gitignore	2023-03-14 23:21:22 +08:00
ddpm.py	docs: delete useless comment	2023-03-14 23:19:21 +08:00
README.md	docs: update readme	2023-03-14 23:33:54 +08:00
sample.py	style: use configparser to manage config file	2023-03-14 23:20:32 +08:00
train.py	style: use configparser to manage config file	2023-03-14 23:20:32 +08:00
training.ini	style: use configparser to manage config file	2023-03-14 23:20:32 +08:00
unet.py	fix: add dropout layer into unet model	2023-03-14 18:18:13 +08:00

DDPM

A Simple implementation of DDPM model in PyTorch.

It's just for fun, the Unet model does not include attention, normalization, etc.

Traning

Before training, please set up the config.ini file:

[unet]
batch_size = 256
time_emb_dim = 128
device = cuda
epoch_num = 500
learning_rate = 1e-4

[ddpm]
iteration = 500

To start training, run:

$ python train.py

To generate 16 pictures, run the following command:

The pictures will be output to the ./output directory.

$ python sample 16