Study 01 · Scene to SimSee all studies

PPISP study on a 3-camera capture

We put NVIDIA's PPISP through one clean test on our own capture: same scene, same 30,000 iterations, one switch. Image quality rose by 9.5 dB.

The same scene side by side: PPISP off on the left, on on the right

Camera correction off and on in this study.

PPISP is NVIDIA's fix for the lighting and colour drift that turns a clean multi-camera capture into mush. It sits on top of 3DGS, 3DGUT and NeRF pipelines.

With it off, the sculpture and the water dissolve into haze; with it on, edges hold and colour stays true.

PPISP off: a hazy, washed-out blurPPISP on: sharp and colour-true

Study results: image quality in three numbers.

+9.5 dB

PSNR with PPISP on, against off

30k

Training iterations in both runs

3

Real cameras on one scene

Why image quality matters for robot training.

Our Scene to Sim pipeline turns real captures into the 3D places where robots learn a skill before hardware moves. A blurry scene gives geometry the robot cannot trust and textures the policy cannot read. Other methods tested in this line of work: 3DGRUT, 2DGS and GOF.

The reconstruction with PPISP on, sharp across the scene

What one switch showed

One switch, the same capture and the same budget: 9.5 dB is the whole difference PPISP made.