VGGT + SAM
Unsuccessful execution
Banana
First-person video
3D geometry · VGGT point cloud
Third-person video
01
Interactive object results
Ground-truth geometry provides a direct reference for shape, appearance, and spatial arrangement.
02
Side-by-side streaming comparison
Long monocular streams reveal how accumulated evidence improves reconstruction over a single-view generation baseline.
03
Downstream robotic manipulation
Each execution record pairs eye-in-hand input, reconstructed 3D geometry, and fixed third-person verification.
Streaming3D
Successful execution
Banana
First-person video
3D geometry · Streaming3D mesh
Third-person video
VGGT + SAM
Unsuccessful execution
Red pepper
First-person video
3D geometry · VGGT point cloud
Third-person video
Streaming3D
Successful execution
Red pepper
First-person video
3D geometry · Streaming3D mesh
Third-person video