Spaces:
Running
Running
| title: PPISP Scene→Sim Ablation | |
| emoji: 🔬 | |
| colorFrom: green | |
| colorTo: gray | |
| sdk: static | |
| pinned: true | |
| license: mit | |
| short_description: Drag-compare NVIDIA PPISP off vs on (3-camera) | |
| # PPISP on a Real 3-Camera Capture — OFF vs ON | |
| An interactive drag-to-compare of **NVIDIA's PPISP** run inside Forenly AI's | |
| **Scene→Sim** pipeline. Same capture, same 30k iterations, one toggle: | |
| - **OFF** → the reconstruction collapses into a hazy, washed-out blur | |
| - **ON** → sharp, colour-correct and stable (**+9.5 dB PSNR**) | |
| PPISP fixes the photometric artifacts that wreck radiance-field and Gaussian-splatting | |
| reconstructions. In our Scene→Sim pipeline we turn real multi-camera captures into | |
| simulation-ready 3D scenes where humanoid robots learn their skills — so every dB of | |
| reconstruction fidelity is a better training ground. | |
| Methods benchmarked in this line of work: **PPISP · 3DGRUT · 2DGS · GOF**. | |
| **Forenly AI — The Skill Layer for Humanoids.** → https://forenly.ai/lab-study/ppisp | |
| Dataset: [ForenlyAI/ppisp-ablation](https://huggingface.co/datasets/ForenlyAI/ppisp-ablation) | |