知识卡片

普通读书笔记卡 · 9081.d · 生成模型:GPT-6

合成数据价值不只在逼真度

理解等级:中级

Flying Chairs通过背景和椅子图像的变换生成大量训练对,展示外观不必高度逼真也能提供有用监督。其作用依赖运动与纹理覆盖以及评估条件,不证明真实数据没有价值,或任意简单合成都能迁移到目标任务。

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

      The largest existing datasets are synthetic 3D scenes with moving objects that can be rendered, which will give us perfect ground truth data to train the regression function. There are several examples of existing datasets such as this, like the Middelbury dataset [425], which contains six real and synthetic sequences with ground truth optical flow. The optical flow for the real sequences was obtained by

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第49章《学习运动估计》(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/c049.xhtml) - 结论依据:摘录中关于合成数据价值不只在逼真度的定义、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:The largest existing datasets are synthetic 3D scenes with moving objects that can be rendered, which will give us perfect ground truth data to train the regression function. There are several examples of existing datasets such as this, like the Middelbury dataset [425], which contains six real and synthetic sequences with ground truth optical flow. The optical flow for the real sequences was obtained by