Difficulties in Regenerating Results with WordStylist: Styled Verbatim Handwritten Text Generation using Latent Diffusion Models

[AniketGurav]

I am currently attempting to regenerate the results using your code for WordStylist: Styled Verbatim Handwritten Text Generation with Latent Diffusion Models. However, I have encountered difficulties in doing so. I have trained the model for 900 epochs without making any changes to the parameters. Unfortunately, the visual results after 900 epochs are not correct (attached ).
can you please provide info about how many epochs you have trained the model.

[koninik]

I added the pre-processing of the images (which I believe is the issue), so please have a look at that and let me know if you get better results.

[AniketGurav]

Hi thanks for update.
I am able to train the model and get the visually correct results after applying preprocessing.