Is the model you use actually BeatGANsAutoencModel and it has no cross attention?

ankanbhunia / PIDM

Person Image Synthesis via Denoising Diffusion Model (CVPR 2023)

https://ankanbhunia.github.io/PIDM

MIT License

481 stars 62 forks source link

Is the model you use actually BeatGANsAutoencModel and it has no cross attention? #68

Closed safeanimal closed 2 days ago

safeanimal commented 2 days ago

Am I wrong? I exmained the code and found that it actually uses BeatGANsAutoencModel, I don't see any cross attention in it, it's different from the method you use in the paper.

ankanbhunia commented 2 days ago

see https://github.com/ankanbhunia/PIDM/blob/e4f1d880772072ee464a83240d0374bf466b6772/models/blocks.py#L471

Follow the code: AttentionBlock -> ResBlock -> BeatGANsAutoencModel

safeanimal commented 2 days ago

see

https://github.com/ankanbhunia/PIDM/blob/e4f1d880772072ee464a83240d0374bf466b6772/models/blocks.py#L471

Follow the code: AttentionBlock -> ResBlock -> BeatGANsAutoencModel

ok, thank you, I mistook AttentionBlock as a self-attention.