Skip to content

doc:refine Artimuse overview wording and sample reference - #162

Closed
Kylie-dot-s wants to merge 1 commit into
MigoXLab:devfrom
Kylie-dot-s:docs/artimuse-md-tweak
Closed

doc:refine Artimuse overview wording and sample reference#162
Kylie-dot-s wants to merge 1 commit into
MigoXLab:devfrom
Kylie-dot-s:docs/artimuse-md-tweak

Conversation

@Kylie-dot-s

Copy link
Copy Markdown
Contributor

No description provided.

Comment thread docs/artimuse.md
`RuleImageArtimuse` 是一个基于 Artimuse API 的图像质量评估规则类,用于判断输入图像的质量是否达到合格标准。该规则通过调用 Artimuse 服务的图像评估接口,获取图像的总体评分和各个方面的详细评估结果。
RuleImageArtimuse 基于 ArtiMuse 服务,面向 AIGC 与创意生产的质量把关与可解释诊断。不同于只给分数或只给文本的方案,ArtiMuse 同时给出整体分与专家级细粒度理解。细粒度理解涵盖八个维度:构图设计、视觉元素、技术执行、想象创意、主题传达、情感反应、整体完形、综合评价。这样既能完成量化筛选,也能给出可直接落地的改进依据。

在我们的本地评估样例20250903_203109_deb630bc 中,整体得分为 75 分,Good 与 Bad 的比例约为七比三。维度解释显示技术执行、构图与视觉元素较为稳健,而想象创意往往是主要扣分来源。这与论文提出的联合评分和专家级理解目标一致:不仅判断好坏,更揭示原因,使结果可以反哺提示词、光线与叙事设置,形成生成、评估与再生成的闭环。

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

“本地评估样例20250903_203109_deb630bc 中”
介绍这个样例是怎么获取的

Comment thread docs/artimuse.md
Comment thread docs/artimuse.md
3. **返回评估结果**
- 根据总体评分判断图像质量是否合格
- 返回包含详细评估信息的 `ModelRes` 对象
1. 创建评估任务:向 ArtiMuse 发送包含 URL 与 style 的请求,创建阶段超时 30 秒

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

看最新代码RuleImageArtimuse

Comment thread docs/artimuse.md
# "threshold": 6.0,
# "style": 1, # 1 表示专业风格,2 表示批评性风格
# "create_timeout_sec": 30,
# "poll_max_tries": 5,

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

没有这些参数

@shijinpjlab shijinpjlab closed this Sep 4, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants