知识卡片

普通读书笔记卡 · 8999 · 生成模型:GPT-6

图像翻译统一多种输出任务

理解等级:中级

颜色图、深度图和语义标签图都可按空间网格表示,因此可以共同作为图像到图像映射的输出。统一张量接口不意味着损失、合法值或不确定性相同:类别标签、度量深度和颜色需要各自合适的分布与评估。

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

      Image-to-image problems are mapping problems where the input is an image and the output is also an image, where we will here think of an image as any array of size N × M × C (height by width by number of channels). These problems are very common in computer vision. Examples include colorization, next frame prediction, depth map estimation, and semantic segmentation (a per-pixel

    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:摘录中关于图像翻译统一多种输出任务的定义、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:Image-to-image problems are mapping problems where the input is an image and the output is also an image, where we will here think of an image as any array of size N × M × C (height by width by number of channels). These problems are very common in computer vision. Examples include colorization, next frame prediction, depth map estimation, and semantic segmentation (a per-pixel