The simple implenmentation of DDPM cuda1.11.0 RTX3090 This model use Cifar10 datasets:https://www.kaggle.com/datasets/joaopauloschuler/cifar10-64x64-resized-via-cai-super-resolution checkpoints:https://www.kaggle.com/datasets/samuelngnjust/ddpm-checkpoint-ema
The structure of network is Unet_conditional which is:
Unet结构Input
|
DoubleConv
|
Down -> SelfAttention
|
DoubleConv
|
Down -> SelfAttention
|
DoubleConv
|
Down -> SelfAttention
|
|
DoubleConv
|
Up -> SelfAttention
|
DoubleConv
|
Up -> SelfAttention
|
DoubleConv
|
Up -> SelfAttention
|
|
Output
Using
!python DDPM_conditional.py"
to train model and generate images