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]
Click here to download "TSUNAMI" dataset (about 15.8MBytes).
Send an email request to the followoing address to download "GSV" dataset.
"TSUNAMI" consists of one hundred panoramic image pairs of scenes in tsunami-damaged areas of Japan.
"GSV" consists of one hundred panoramic image pairs of Google Street View.
The contents of GSV are same as TSUNAMI dataset.
"TSUNAMI" consists of one hundred panoramic image pairs of scenes in tsunami-damaged areas of Japan. "GSV consists of one hundred panoramic image pairs of Google Street View. The size of these images is 224 × 1024 pixels.
For each of them, we hand-labeled the ground truth of scene changes. It is given in the form of binary image of the same size as the input pair of images. The binary value at each pixel indicates that a change has occurred at the corresponding scene point on the paired images. We defined the scene changes to be detected as 2D changes of surfaces of objects (e.g., changes of the advertising board) and 3D, structural changes (e.g., emergence/vanishing of buildings and cars). The changes due to differences in illumination and photographing condition and those of the sky and the ground are excluded, such as changes due to specular reflection on building windows and changes of cloud and signs on the road surface.
All image pairs have ground truths of temporal scene changes, which are manually obtained by ourselves. They are stored in "ground_truth/*.bmp."
TSUNAMI or GSV
| --t0 // *.jpg
| --t1 // *.jpg
| --ground_truth // *.bmp
We welcome your questions, comments and suggestions. Please send them to sakurada"at"vision.is.tohoku.ac.jp or okatani"at"vision.is.tohoku.ac.jp
Ken Sakurada and Takayuki Okatani
Tohoku University, Japan