关于纯图像的得分评估

[zcdliuwei]

您好，感谢您开源如此重要的AIGC图文得分偏好模型和方法。
我看您readme文件中的如下命令：
rewards = model.score("", ["<img1_obj_or_path>", "<img2_obj_or_path>", ...])
是在评估同一个prompt生成的不同图像的得分偏好。如果我只有一堆图文对，即 [[prompt1, image1], [prompt2, image2], ...[promptN, imageN]]，他们之间是一一对应，而不是一对多的关系，请问，这种情况下我应该如何使用您的模型对这些图文对的美学、人类偏好进行得分排序？
期待您的回复

[hkunzhe]

Have a try at score. However, from my experience, these metrics are best used to compare multiple images generated by the same prompt. This is due to the training dataset and training objective.

[xujz18]

Thanks a lot to @hkunzhe for sharing and explaining!

[zcdliuwei]

Thank you for your reply
However, I see that prompt still needs to be input in the score function：