xrenaa commited on
Commit
f156376
·
verified ·
1 Parent(s): 6bf30e4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -0
README.md CHANGED
@@ -4,6 +4,8 @@
4
 
5
  # Lyra: Generative 3D Scene Reconstruction via Video Diffusion Model Self-Distillation
6
 
 
 
7
  [Sherwin Bahmani](https://sherwinbahmani.github.io/),
8
  [Tianchang Shen](https://www.cs.toronto.edu/~shenti11/),
9
  [Jiawei Ren](https://jiawei-ren.github.io/),
@@ -18,6 +20,7 @@
18
  [Jun Gao](https://www.cs.toronto.edu/~jungao/),
19
  [Xuanchi Ren](https://xuanchiren.com/) <br>
20
 
 
21
  ### Description:
22
  Lyra is a 3D / 4D feed-forward 3D Gaussian Splatting (3DGS) reconstruction model. We achieve this by distilling a pre-trained video diffusion model into a feed-forward 3D Gaussian Splatting (3DGS) generator. Lyra circumvents the need for 3D datasets or model fine-tuning by leveraging the inherent 3D consistency of video outputs to align 2D renderings with a 3DGS decoder. By using the generated synthetic multi-view data, we train a decoder to operate directly in the video model's latent space to produce 3D Gaussians. This framework enables real-time rendering and establishes a new state of the art for 3D / 4D scene generation, supporting both single-view and video inputs.
23
 
 
4
 
5
  # Lyra: Generative 3D Scene Reconstruction via Video Diffusion Model Self-Distillation
6
 
7
+ **[Paper](), [Project Page](https://research.nvidia.com/labs/toronto-ai/lyra/)**
8
+
9
  [Sherwin Bahmani](https://sherwinbahmani.github.io/),
10
  [Tianchang Shen](https://www.cs.toronto.edu/~shenti11/),
11
  [Jiawei Ren](https://jiawei-ren.github.io/),
 
20
  [Jun Gao](https://www.cs.toronto.edu/~jungao/),
21
  [Xuanchi Ren](https://xuanchiren.com/) <br>
22
 
23
+
24
  ### Description:
25
  Lyra is a 3D / 4D feed-forward 3D Gaussian Splatting (3DGS) reconstruction model. We achieve this by distilling a pre-trained video diffusion model into a feed-forward 3D Gaussian Splatting (3DGS) generator. Lyra circumvents the need for 3D datasets or model fine-tuning by leveraging the inherent 3D consistency of video outputs to align 2D renderings with a 3DGS decoder. By using the generated synthetic multi-view data, we train a decoder to operate directly in the video model's latent space to produce 3D Gaussians. This framework enables real-time rendering and establishes a new state of the art for 3D / 4D scene generation, supporting both single-view and video inputs.
26