Spaces:

---

ruili3

/

LaRI

like 5

Running

Dear HF team,

LaRI is a novel method for modeling occluded and unseen 3D geometry from a single image using layered point maps. Unlike standard depth estimation, LaRI captures multiple surfaces per ray and unifies object—and scene-level reasoning in a single feed-forward pass.

The demo visualizes multi-layer 3D predictions and supports interactive comparisons with existing depth models. It also showcases generalization to both synthetic and real-world data. A GPU is essential to maintain real-time performance for user interaction and rendering.

We therefore request a free GPU allocation to allow for a smoother user interaction to promote this new line of research with HuggingFace. Thank you for considering our request!

Best,
Rui