Details on quantitative results

[HoiM]

Thank you for your work. When calculating CLIP-I and CLIP-T, did you use text prompt as an input into the model? Is the image prompt the only input?

[xiaohu2015]

coco dataset have text captions

[HoiM]

coco dataset have text captions

I know it has captions. What I plan to do is to conduct same evaluation with my own IP-Adapter.

I don't know if I should use text prompt when the model generates image samples.

[greasebig]

@xiaohu2015 hi, some questions about paper.

1.in your paper, says "we generate 4 images conditioned on the image prompt for each sample in the dataset, resulting in total 20,000 generated images for each method", did you generate 4 different images by different seed? or different captions?

2.when you got quantitative results, do your generated images have the same size with each validate coco image? (like width=600, height=400, same with coco eval annotations)

Would you be willing to provide the code for validating the quantitative indicators to other researchers?