知识卡片
语义分离图要结合训练目标读
CLIP训练的视觉模型在靠近输出层处呈现较清楚的语义类分组,与其训练目标有关。该图是特定模型、数据及投影下的观察,不证明所有网络每层都单调提高可分性,也不能只凭好看的投影取代下游分类或迁移评估。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
this network has been trained to output a direct representation of semantics (in particular, this net was trained with contrastive language-image pre-training [CLIP [394]], which is a method for learning semantic representations that we will learn about in section 51.3).
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第13章《作为分布变换器的神经网络》(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/c013.xhtml)
- 结论依据:摘录中关于语义分离图要结合训练目标读的定义、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:this network has been trained to output a direct representation of semantics (in particular, this net was trained with contrastive language-image pre-training [CLIP [394]], which is a method for learning semantic representations that we will learn about in section 51.3).