dcharatan / pixelsplat

[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
http://davidcharatan.com/pixelsplat/
MIT License
830 stars 56 forks source link

Question about generalizable #44

Closed SuzyZhao0417 closed 6 months ago

SuzyZhao0417 commented 6 months ago

Hello! Thank you for your excellent work! Regarding generalization, I have some questions. Could you please clarify whether your model achieves scale generalization within the same scene, or if it utilizes pre-trained models to achieve generalization on unseen scenes? Looking for your reply! Best wishes!

dcharatan commented 6 months ago

Can you explain what you mean by achieving scale generalization within the same scene? You can overfit to a single scene (see #40), but during overfitting of any NeRF-like model, scale ambiguity isn't a concern. The model is designed to not require metric-scale camera poses during generalization to new scenes.