1 oral and 1 poster paper accepted to ICCV 2021

We are excited to announce that two papers were accepted to ICCV 2021. These are results from the joint research between Matsushita Laboratory at Osaka University and LINE Corporation.

  • Feiran Li, Kent Fujiwara, Fumio Okura, and Yasuyuki Matsushita, “Generalized Shuffled Linear Regression” (Oral)
  • Feiran Li, Kent Fujiwara, Fumio Okura, and Yasuyuki Matsushita, “A Closer Look at Rotation-invariant Deep Point Cloud Analysis”

In the first paper, we focused on the problem of Shuffled Linear Regression (SLR). SLR is a special case of linear regression where the correspondence between covariates and responses are unknown. However, such scenarios seldom exist in practice because of missing data and outliers. We define Generalized Shuffled Linear Regression (GSLR), a more general case of SLR where the correspondence is unknown and there is existence of missing data. We propose a simple method to solve GSLR with a detailed analysis regarding optimization.

In the second paper, we reviewed the current works regarding pose-invariant deep point cloud analysis. We discovered that methods generally attempt to align point clouds using PCA. However, by using all the combinations of the principle axes rather than fixing them based on the order of the eigenvalues, we discovered that this data augmentation itself achieves higher accuracy than most current complex methods. We further proposed an attachable module for neural networks that selects the optimal combination of the axes for better accuracy.

The papers will be presented at ICCV (Oct. 11-17, 2021). Papers will be released soon.

Kent Fujiwara
Kent Fujiwara
Research Scientist

Computer Vision Research Scientist