Chaofeng Chen

transformers 4.37.2 is required by the `qalign` metric. Feel free to use latest transformers if you are not using `qalign`.

Actually, there is a similar metric in our toolbox, `topiq_nr-face`. You may use it currently.

Please consider handling these cases manually in Python using `try...except` blocks. Returning `NA` may lead to incorrect or misleading error messages in reports.

I have evaluated the performance of `topiq_nr-face` on the [DSL-FIQA](https://github.com/DSL-FIQA/DSL-FIQA) test set. Despite being trained on GFIQA-20K, `topiq_nr-face` demonstrates results comparable to DSL-FIQA. Given that DSL-FIQA requires additional landmark detection,...

For assistance with writing Python code, please consider reaching out to ChatGPT or Claude; they can provide the support you need. For other issues, they may be outside the scope...

I directly used pytorch-FID with CelebA-HQ-Test as reference set and did not use extra codes

You might want to consider modifying the following codes: https://github.com/chaofengc/FeMaSR/blob/1ca4295a2430057407f32d0d3fba70270ca68ce7/options/train_FeMaSR_HQ_pretrain_stage.yml#L18 https://github.com/chaofengc/FeMaSR/blob/1ca4295a2430057407f32d0d3fba70270ca68ce7/options/train_FeMaSR_HQ_pretrain_stage.yml#L43 > [!CAUTION] The provided code is **NOT optimized** for $512\times512$ images. If you intend to work with this resolution,...

Thank you for your email. I believe an open discussion here would be more beneficial, as it could also assist others who may have similar questions. --- The pretrained model...

Thank you for your feedback and suggestion, Regarding the document, it takes time to write detailed documents and I am sorry that I am not available to do that at...