pub fn pictured(query: &str) -> boolExpand description
Whether a query describes something seen, which is what the image list is for (0.42.0, G13).
A third reading of the same text, beside shape_of and code_shaped
for the same reason they sit together. CLIP cannot be the judge: it reads
the words in a screenshot, so a question naming the project scores the
project’s own screenshots like a description of them. Measured on flask,
“how does flask dispatch a request to a view” put flaskr_index.png at a
cosine of 0.355 — the right picture for “a green triangle” scores 0.359 —
and no floor or margin over a caption true of any picture told the two
apart. The query’s own words do: a picture word (screenshot, diagram,
logo), a colour or a shape. A crude list a reader can predict, as with
the shape, rather than a judgement nobody can.