diffusion_models.models.openai_unetΒΆ
Classes
|
An attention block that allows spatial positions to attend to each other. |
|
A downsampling layer with an optional convolution. |
|
A module which performs QKV attention. |
|
A residual block that can optionally change the number of channels. |
|
A UNetModel that performs super-resolution. |
|
Any module where forward() takes timestep embeddings as a second argument. |
|
A sequential module that passes timestep embeddings to the children that support it as an extra input. |
|
The full UNet model with attention and timestep embedding. |
|
An upsampling layer with an optional convolution. |