← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9011.b · 生成模型:GPT-6

图注优先辨清提示任务

理解等级:中级

图 37.6 展示对象、场景、美学三种视觉提示与相应输出;同一输入经不同提示用于不同任务。邻近正文却说狗的品种与表情,与图不一致。保留图中可见任务作为示例,核心机制是优化输入信号来调节固定模型,而不是据此推断所有提示都通用。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Figure 37.6: A prompt can be made out of pixels: a prompt image (which looks like a border of noise here) is added to the input image in order to change the model's behavior. Three different prompts are shown, one that adapts the model to perform scene classification, a second for aesthetics classification, and a third for object classification. Modified from [28].

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第37章《迁移学习与适配》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c037.xhtml) - 结论依据:摘录中关于图注优先辨清提示任务的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Figure 37.6: A prompt can be made out of pixels: a prompt image (which looks like a border of noise here) is added to the input image in order to change the model's behavior. Three different prompts are shown, one that adapts the model to perform scene classification, a second for aesthetics classification, and a third for object classification. Modified from [28].