CLIP Score Evaluation without Pre-processing.

I am referring to [Evaluating Diffusion Models](https://huggingface.co/docs/diffusers/main/en/conceptual/evaluation), specifically the quantitative evaluation using CLIP score example.

We have images of shape (6, 512, 512, 3).

CLIP score is calculated using `"openai/clip-vit-base-patch16"`.

However, as far as I can tell, the images are not pre-processed to match the format that `"openai/clip-vit-base-patch16"` was trained on (e.g., images of size 224x224 pixels).

Should the images have been processed before or can we still reliably use the CLIP score with the images in their original format?

Please let me know if I have overlooked or am misunderstanding something. Thanks!

Start with the quantitative CLIP score example in the Evaluating Diffusion Models documentation and inspect how the (6, 512, 512, 3) images are passed to openai/clip-vit-base-patch16. Compare that path with the model's expected image inputs. Done means the documentation clearly states whether preprocessing is applied or required, and the example is corrected if needed.

Scritto dal modello di indicizzazione a partire dal testo della issue.