Are the evaluation prompts / inference test set available?

[MapleAndJoker]

Hi, thanks for the great work and for releasing the code!

I have a question regarding the evaluation protocol used in the generation experiments.
In Section 4, the paper mentions that the quantitative evaluation is conducted on
“132 prompts × 5 Valence × 5 Arousal values” (3,300 images per method).

I was wondering:

1. Are the 132 prompts used for inference / evaluation publicly available?
2. If so, would it be possible to release them (or provide a brief description of how they were constructed)?
3. Otherwise, were these prompts manually curated, or sampled from a specific dataset?

Having access to the evaluation prompts would be very helpful for reproducing and fairly comparing against your method.

Thanks again for the excellent work!

[dangsq]

The prompts are generated by GPT-4 using the following instruction:
Generate a set of {n} text-to-image prompts. Each prompt should be a single sentence with a clear subject–verb–object structure, depicting a scene with diverse settings and atmospheres.