[CVPR 2024 Oral, Best Paper Runner-Up] Code for "pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction" by David Charatan, Sizhe Lester Li, Andrea Tagliasacchi, and Vincent Sitzmann
Hello, I don’t have a deep understanding of the transformer structure, but I still want to ask if there is a naming error before and after. The feature of epipolar sampling was first named Q, but it was actually used as KV?
The code originally had the q and kv variable names flipped. That got fixed in this commit. Make sure you pull the latest code, since it fixes this issue and a few other small issues and has slightly better performance!
Hello, I don’t have a deep understanding of the transformer structure, but I still want to ask if there is a naming error before and after. The feature of epipolar sampling was first named Q, but it was actually used as KV?