Low-Rank Tensor Completion with Total Variation for Visual Data Inpainting
Authors: Xutao Li, Yunming Ye, Xiaofei Xu
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on color image and medical image inpainting tasks show the effectiveness and superiority of the two methods against state-of-the-art competitors. |
| Researcher Affiliation | Academia | Xutao Li, Yunming Ye, Xiaofei Xu Shenzhen Key Laboratory of Internet Information Collaboration, Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen, 518055, China Emails: lixutao@hitsz.edu.cn; yym@hitsz.edu.cn;xiaofei@hit.edu.cn |
| Pseudocode | Yes | Algorithm 1: LRTC-TV-I |
| Open Source Code | Yes | Our codes are available at https://sites.google.com/site/xutaoli2014 |
| Open Datasets | Yes | We show in Figure 1 the ground-truth of eight images used for the experiment. The BRAINIX data set1 is utilized, and we build a 288-by-288-by-22 tensor. 1http://www.osirix-viewer.com/datasets/ |
| Dataset Splits | No | To test the inpainting performance, we mask off 60%, 65%, 70%, 75%, 80%, 85%, 90% and 95% of entries in each image randomly, and regard them as missing values. The remaining points, making up the incomplete tensor Y, are leveraged to recover the original tensor. |
| Hardware Specification | No | No specific hardware details (such as GPU or CPU models, or memory) are provided for the experimental setup. |
| Software Dependencies | No | The paper mentions the use of the ADMM framework and describes algorithmic steps, but does not provide specific software names with version numbers (e.g., Python, PyTorch, specific numerical libraries) that would be needed for replication. |
| Experiment Setup | Yes | Parameter Settings. In both inpainting tasks, tensors of third order are considered, whose first two modes stand for spatial dimensions and third mode denotes channel information. Hence, we set β1 = β2 = 1 and β3 = 0 for LRTC-TV-I and LRTC-TV-II in both tasks. In LRTC-TV-I, the parameter λ is set to be 2.0 × 10−2. In LRTC-TV-II, we utilize λ1 = 5.0 × 10−1 and λ2 = 1.0 × 103. For both methods, we set K = 300, μ = 1.1 and ρ1 = ρ2 = ρ3(= ρ4) = 1.0 × 10−2. The parameters of baseline approaches are specified optimally by following the papers in which these approaches were developed and proposed. |