How is the reward mode designed?

[EthanLeo-LYX]

In the paper Section 4.2 Reward, you said that you developed a reward model to assess the performance by calculating the similarity between the final UI page and the object UI page. I wonder how the reward model is designed and trained. And would the reward model be released?

[csdaa]

这个项目视乎被抛弃了，换老外的吧

[csdaa]

这里有一篇他们的论文，也许能帮到你 https://arxiv.org/pdf/2312.13771.pdf