Third-Party Notices
This code includes adaptations of the following research implementations. Their licenses and attribution requirements must be reviewed before assigning a license to this repository.
DiT
- Project: Scalable Diffusion Models with Transformers
- Source: https://github.com/facebookresearch/DiT
- Used concepts/code: timestep embedding, adaLN-Zero transformer blocks, initialization, and diffusion-model structure
- Copyright: Meta Platforms, Inc. and affiliates
OpenAI diffusion implementations
- Projects: improved-diffusion, guided-diffusion, and GLIDE
- Sources:
- Used code: Gaussian diffusion schedules, training losses, timestep respacing, and DDIM sampling
timm
- Project: PyTorch Image Models
- Source: https://github.com/huggingface/pytorch-image-models
- Used components: Transformer attention and MLP modules