Request for Training Code (Soft Token / Prompt Learning Pipeline)

[jeje910]

Hi, thank you for the great work and for releasing the pretrained weights of Diff-II.

I am currently conducting research on fine-grained text-to-image tuning and would like to reproduce your results more faithfully. In particular, I am very interested in the soft token / soft prompt learning pipeline used during training.

Would it be possible to release the corresponding training code or a minimal example of the soft token optimization process? Having access to the training pipeline would greatly help the community understand and extend Diff-II.

[scuwyh2000]

Hi, thank you for the great work and for releasing the pretrained weights of Diff-II.

I am currently conducting research on fine-grained text-to-image tuning and would like to reproduce your results more faithfully. In particular, I am very interested in the soft token / soft prompt learning pipeline used during training.

Would it be possible to release the corresponding training code or a minimal example of the soft token optimization process? Having access to the training pipeline would greatly help the community understand and extend Diff-II.

Hi,

As indicated in Appendix A.4, for the Category concept learning, we followed the implementations of ``Enhance
image classification via inter-class image mixup with diffusion model.''

[jeje910]

Thank you for your quick reply! I'll dig on it