48 lines
986 B
Markdown
48 lines
986 B
Markdown
# DDPM
|
|
|
|
A Simple implementation of DDPM model in PyTorch.
|
|
|
|
It's just for fun, the Unet model does not include attention, normalization, etc.
|
|
|
|
|
|
|
|
|
|
## Classifier Guidence DDPM
|
|
ref: Diffusion Models Beat GANs on Image Synthesis (https://arxiv.org/abs/2105.05233)
|
|
- generate "7"
|
|
- 
|
|
- generate "2"
|
|
- 
|
|
|
|
## Traning
|
|
|
|
Before training, please set up the config.ini file:
|
|
|
|
```ini
|
|
[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
|
|
```
|
|
|
|
## Sampling
|
|
|
|
To generate 16 pictures, run the following command:
|
|
|
|
The pictures will be output to the `./output` directory.
|
|
|
|
```
|
|
$ python sample 16 # unconditional
|
|
$ python sample 16 7 # condiditional, want to generate "7" pictures
|
|
``` |