Download

This dataset consists of two subsets, named "TSUNAMI" and "GSV." We make them publicly available for the researchers who are interested in the problem of the image-based detection of temporal scene changes. Although we own its copyright, you can freely use it for research purposes. We request that you cite the following paper if you publish research results utilizing these data:

Ken Sakurada and Takayuki Okatani, Change Detection from a Street Image Pair using CNN Features and Superpixel Segmentation, Proceedings of British Machine Vision Conference (BMVC), 2015. [pdf]

Read more...


This dataset contains the image sequences of city streets captured by a vehicle-mounted camera at two different time points. We make them publicly available for the researchers who are interested in the problem of the image-based detection of temporal changes of 3D scene structures. Although we own its copyright, you can freely use it for research purposes. We request that you cite the following paper if you publish research results utilizing these data:

Ken Sakurada, Takayuki Okatani, Koichiro Deguchi, Detecting Changes in 3D Structure of a Scene from Multi-view Images Captured by a Vehicle-mounted Camera, Proc. Computer Vision and Pattern Recognition, 2013.

Read more...


Important note: We have realized that the code distributed below is a wrong version (3/28/2014). The MATLAB function damped_wiberg() does not achieve the performance we reported in our ICCV paper. You can download the correct version from here, which is a zip file containing a single MATLAB function that replaces damped_wiberg() in the packages below. We are sorry for any incovenience caused.

---

You can download the MATLAB code of the damped Wiberg method from here.

The zip file contains a MATLAB function that factorizes a given matrix Y into U*V', a product of two smaller matrices of rank r. To be specific, it minimizes ||H \odot (Y - U*V')||_2 for specified data Y, H, and r, where Y is the matrix to be factorized, H is the binary matrix indicating existing (1) and missing (0) components of Y, and r is the number of columns of U and V. For more detail, see the source code.

For the details of the implementation, please see our paper below.

[1] Takayuki Okatani, Takahiro Yoshida, Koichiro Deguchi: Efficient algorithm for low-rank matrix factorization with missing components and performance comparison of latest algorithms. Proc, ICCV 2011: 842-849. [pdf]