Skip to content

Implement text decoding submodule to ME2-VD #16

@reesekneeland

Description

@reesekneeland

Brain Diffuser decodes both the CLIP Text vectors for the image captions as well as the CLIP Image vectors for the images themselves. Brain Diffuser is the best performer on imagery, thus, one potential avenue for improving imagery recons might be incorporating text decoding for dual guidance in versatile diffusion. This issue is to implement a decoding submodule for the caption info, potentially with it's own diffusion prior if GPU memory allows.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions