← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9005.c · 生成模型:GPT-6

域随机化改变仿真参数

理解等级:中级

数据增强变换已有样本,域随机化则改变模拟器的照明等参数来生成不同经验。两者都尝试扩大训练条件的覆盖。要围绕目标环境可能变化的因素设定采样范围;仿真内部变得多样,不代表已覆盖现实中未建模的因素。

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

      Sometimes our data is not sampled from a static dataset but instead is generated by a simulator. In this setting we may apply an idea analogous to data augmentation called domain randomization. Here, rather than adding random transformations to points in a dataset, we instead randomize the parameters of the simulator, so that each interaction with the simulation will look slightly different, for example, the

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第36章《鲁棒性与通用性训练》(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/c036.xhtml) - 结论依据:摘录中关于域随机化改变仿真参数的定义、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:Sometimes our data is not sampled from a static dataset but instead is generated by a simulator. In this setting we may apply an idea analogous to data augmentation called domain randomization. Here, rather than adding random transformations to points in a dataset, we instead randomize the parameters of the simulator, so that each interaction with the simulation will look slightly different, for example, the