Request inference instruction: unable to reproduce colorization results as shown in the paper

[qiuqianru]

Thank you for your excellent work and the impressive results presented in your paper.

I have attempted to train and fine-tune the model using the provided scripts (process/train_animediffusion.py and ft_animediffusion.py), with 300 epochs for training and 10 epochs for fine-tuning. The colorization results look good when the line art is identical to the reference image. However, when using a different line art, the model seems to struggle to achieve similar results. All line art images are preprocessed by XDoG function.

Could you provide some guidance on how to improve the colorization performance with different line art inputs, or share any tips for using the model?

Thank you very much for your assistance!

[SystemErrorWang]

Hello @qiuqianru, would you please tell me how did you manage to start the training? have you found the ADF300_10.pth file?