1: \begin{abstract}

2: Transform Invariant Low-rank Textures (TILT) is a novel and

3: powerful tool that can effectively rectify a rich class of

4: low-rank textures in 3D scenes from 2D images despite significant

5: deformation and corruption. The existing algorithm for solving

6: TILT is based on the alternating direction method (ADM). It

7: suffers from high computational cost and is not theoretically

8: guaranteed to converge to a correct solution to the inner loop. In

9: this paper, we propose a novel algorithm to speed up solving TILT,

10: with guaranteed convergence for the inner loop. Our method is

11: based on the recently proposed linearized alternating direction

12: method with adaptive penalty (LADMAP). To further reduce

13: computation, warm starts are also introduced to initialize the

14: variables better and cut the cost on singular value decomposition.

15: Extensive experimental results on both synthetic and real data

16: demonstrate that this new algorithm works much more efficiently

17: and robustly than the existing algorithm. It could be at least

18: \emph{five times faster} than the previous method.

19: \keywords{Transform Invariant Low-Rank Texutres\and Linearized

20: Alternating Direction Method with Adaptive Penalty\and Warm

21: Start\and Singular Value Decomposition}

22: \end{abstract}

23: