原书出处
Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)
Figure 51.11: Image-to-text using a small transformer with cross-attention. The edges in gray are cross-attention edges. Many variations of this architecture are possible. For example, each layer of the text encoder could cross-attend to the output image tokens. Or, the image encoder could compress all its input tokens into a single output token that represents the entire photo; this output token would then get cross-attended
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c051.xhtml
EPUB SHA-256:60a7eebc43f5d9a675d776fed3736fbdaa791e55877023f2f8cba016a9b384f8