知识卡片

普通读书笔记卡 · 8998.c · 生成模型:GPT-6

添加条件不保证模型服从条件

理解等级:中级

若条件与训练目标关系弱、网络可忽略该分支或数据存在偏差,加入文字或图像条件并不保证生成结果正确对应它。应在固定随机性下改变条件并检查语义、结构和多样性,区分条件有效性与单张图像外观质量。

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

      An example where we condition on text is shown in figure 34.6. The intuition is that if we give fθ a description of the image as an additional input, then it can do a better job at solving its prediction task. Because of this, fθ will become sensitive to the text command, and if you give different text it will denoise toward a different image—an

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第34章《条件生成模型》(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/c034.xhtml) - 结论依据:摘录中关于添加条件不保证模型服从条件的定义、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:An example where we condition on text is shown in figure 34.6. The intuition is that if we give fθ a description of the image as an additional input, then it can do a better job at solving its prediction task. Because of this, fθ will become sensitive to the text command, and if you give different text it will denoise toward a different image—an