知识卡片
零样本分类比较文本候选
把候选类描述和查询图像编码为单位向量,选相似度最大的文本作为预测,无需为这一组候选再训练模型。所谓零样本通常指目标任务适配,模型此前已经用图文训练;它不保证概念从未出现在预训练数据中。
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
- · 视觉与语言
原书出处 · 1 处
- · 查看原书依据
3. See which sentence embedding is closest to your image embedding; that's the predicted class. These steps are visualized in figure 51.5. The green outlined dot product is the highest, indicating that the query image will be classified as class a, which is defined as “striped red circles.” Figure 51.5: Making a “striped red circles” (class a) versus “polka dotted green square” (class b) classifier
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:摘录中关于零样本分类比较文本候选的定义、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:3. See which sentence embedding is closest to your image embedding; that's the predicted class. These steps are visualized in figure 51.5. The green outlined dot product is the highest, indicating that the query image will be classified as class a, which is defined as “striped red circles.” Figure 51.5: Making a “striped red circles” (class a) versus “polka dotted green square” (class b) classifier