Unable to Reproduce Results with Provided Checkpoints and Local Training

[Shelby-Lv]

Thank you for sharing this valuable project. I appreciate the time and effort put into making this resource available.

I’ve been trying to reproduce the results reported in the paper using the code and checkpoints provided here, but I’ve encountered a few issues:

Checkpoint Quality: It seems there may be an issue with the uploaded checkpoints for Multi_experts and PPAP. The images sampled using these checkpoints yield significantly lower scores than expected, suggesting the checkpoints might not match those used in the paper.

Local Training Results: After completing local training, I observed an improvement in scores. However, the best Inception Score I achieved with the trained Multi_experts model was 67, which is still below the 74 reported in the paper.

Would it be possible to verify the checkpoints or provide guidance on achieving the reported performance? I would greatly appreciate any advice or insights you could share.

Thank you again for making this project available, and I look forward to any suggestions.

Best regards.