2025.06.11 - #38 - FreeTimeGS, PartCrafter, SuperRANSAC, 4DGT, STORM, LocoTouch, SuperLoc #40
Description
Interesting papers
FreeTimeGS
- https://zju3dv.github.io/freetimegs/
- https://www.4dv.ai/viewer/salmon_10s?showdemo=4dv
- 기존의 4D Gaussian Splatting에 비해, Gaussian primitive가 어디서든 나타날 수 있게 설계하고, 또 각각의 Gaussian primitive가 motion function을 가지고 시간에 따른 움직임을 가질 수 있게 설계.
- 데모가 장난이 아님...
PartCrafter
- https://github.com/wgsxm/PartCrafter
- https://wgsxm.github.io/projects/partcrafter/
- Single shot generation. Traditional two-stage generation suffers from accuracy loss (image segmentation -> parts generation)
SuperRANSAC: One RANSAC to Rule Them All
- RANSAC의 왕... Daniel Barath...
- https://arxiv.org/pdf/2506.04803
- https://github.com/danini/superansac
4DGT: Learning a 4D Gaussian Transformer Using Real-World Monocular Videos
- Learning a 4D Gaussian Transformer Using Real-World Monocular Videos
- https://4dgt.github.io
- 4D Gaussian-based Transformer model for dynamic scene reconstruction, trained entirely on real-world monocular posed videos
- Our model processes 64 consecutive posed frames in a rolling-window fashion, predicting consistent 4D Gaussians in the scene.
- Unlike optimization-based methods, 4DGT performs purely feed-forward inference, reducing reconstruction time from hours to seconds and scaling effectively to long video sequences.
STORM
- Spatio-Temporal Reconstruction Model for Large-Scale Outdoor Scenes
- https://github.com/NVlabs/GaussianSTORM
- https://jiawei-yang.github.io/STORM/
- data-driven Transformer architecture that directly infers dynamic 3D scene representations—parameterized by 3D Gaussians and their velocities—in a single forward pass
