E-RayZer: Self-supervised 3D Reconstruction as Spatial Visual Pre-training
E-RayZer, a self-supervised 3D Vision model predicting camera poses and scene geometry as 3D Gaussians.
Upload multi-view images
Drop File Here
- or -
Click to Upload
Upload your images above or pick a curated example below.
Examples
Preprocessed Images
Run Inference
Predicted target views
Gaussian point cloud
Rendered sweep
Download outputs (zip)
Log