Skip to content

Source paper

tttLRM: Test-Time Training for Long Context and Autoregressive 3D Reconstruction

Chen Wang, Hao Tan, Wang Yifan and 6 more

Chen Wang, Hao Tan, Wang Yifan, Zhiqin Chen, Yuheng Liu, Kalyan Sunkavalli, Sai Bi, Lingjie Liu, Yiwei Hu

arXiv 2602.20160 · 20262 upvotes on Hugging Face

Research ideas from this paper (3)

Generated by AI. The paper’s authors did not write these.

AI summary

The paper presents tttLRM, a Large Reconstruction Model that utilizes Test-Time Training (TTT) layers to perform 3D reconstruction from long sequences of images with linear computational complexity. It enables both feedforward and autoregressive (streaming) reconstruction of explicit 3D representations like Gaussian Splats, scaling efficiently to large context windows where traditional Transformers fail.

How it works (AI summary)

The model replaces standard self-attention with Large Chunk Test-Time Training (LaCT) blocks. Input image tokens update the layer's internal 'fast weights' (a compressed latent memory), which are subsequently queried by virtual view tokens to decode explicit 3D parameters (e.g., 3D Gaussian attributes or NeRF triplanes).