seems possible if u focused on just humans tho
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
i want to make one that can stitch together like pictures you took of a national park or something into a rough sketch of certain areas in 3D (tbh google probably has this given their photo data but can’t figure out how to monetize it)
-
COLMAP is what they used for megadepth (rly good results), i think openmvg works ok too
- Još 4 druga odgovora
Novi razgovor -
-
-
there are a ton of these and they almost never work in more general scenarios https://arxiv.org/pdf/1906.06543.pdf …
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
this seems pretty hard i can't think of a good dataset with ground truths. either you'd have to make it or you'd have to render your candidate solutions in the training loop
-
there might be some applications with a lot of identical-looking shit where low-fi results are acceptable the one I can think of is GIS (just assume all buildings are boxes). but (from googling around) the state of the art seems pretty primitive.
- Još 1 odgovor
Novi razgovor -
-
-
classic photogrammetry is pretty decent for this task usually
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
not what ur looking for exactly but look into meshroom (this isn't perfect to do with neuralnets entirely yet)
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.