← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9095.c · 生成模型:GPT-6

图文负例可能语义上也正确

理解等级:中级

批内非配对组合被标成负例,是一种训练组织规则,不能证明它们语义不匹配。相同场景描述或重复图片可形成假负例;批大小、去重与多正例处理会改变训练信号,不应将所有非对角点积都机械压到最小。

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 51.2 visually depicts how CLIP is trained. First we sample a batch of N language-image pairs, (N = 6 in the figure). Next we embed measure the dot product between all these text strings and images using a language encoder ft and an image encoder fℓ, respectively. This produces a set of text embeddings and a set of image embeddings . To compute the

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第51章《视觉与语言》(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/c051.xhtml) - 结论依据:摘录中关于图文负例可能语义上也正确的定义、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 51.2 visually depicts how CLIP is trained. First we sample a batch of N language-image pairs, (N = 6 in the figure). Next we embed measure the dot product between all these text strings and images using a language encoder ft and an image encoder fℓ, respectively. This produces a set of text embeddings and a set of image embeddings . To compute the