L4GM: Large 4D Gaussian Reconstruction Model

TL;DR

L4GM is a fast, scalable 4D reconstruction model that generates animated 3D objects from a single video frame, leveraging a new multiview dataset and temporal self-attention for high-quality, in-the-wild animation.

Contribution

The paper introduces L4GM, the first 4D large reconstruction model capable of single-view video animation, built on a novel multiview dataset and enhanced with temporal self-attention.

Findings

L4GM produces high-quality animated 3D objects from single-view videos.

The model generalizes well to in-the-wild videos despite training on synthetic data.

L4GM operates in a single second per inference, enabling real-time applications.

Abstract

We present L4GM, the first 4D Large Reconstruction Model that produces animated objects from a single-view video input -- in a single feed-forward pass that takes only a second. Key to our success is a novel dataset of multiview videos containing curated, rendered animated objects from Objaverse. This dataset depicts 44K diverse objects with 110K animations rendered in 48 viewpoints, resulting in 12M videos with a total of 300M frames. We keep our L4GM simple for scalability and build directly on top of LGM, a pretrained 3D Large Reconstruction Model that outputs 3D Gaussian ellipsoids from multiview image input. L4GM outputs a per-frame 3D Gaussian Splatting representation from video frames sampled at a low fps and then upsamples the representation to a higher fps to achieve temporal smoothness. We add temporal self-attention layers to the base LGM to help it learn consistency across time, and utilize a per-timestep multiview rendering loss to train the model. The representation is upsampled to a higher framerate by training an interpolation model which produces intermediate 3D Gaussian representations. We showcase that L4GM that is only trained on synthetic data generalizes extremely well on in-the-wild videos, producing high quality animated 3D assets.