ToHoku University Multi-View Stereo (THU-MVS) Datasets

Multi-view 3D reconstruction datasets: "Cat" and "Dog"

"Cat" and "Dog" datasets are aimed to evaluate the accuracy of multi-view 3D reconstruction algorithms\ , and are freely available in this webpage.

The datasets consist of a set of multi-view images and the ground-truth mesh model. The target objects are figurines of a cat and a dog. The images are taken with a grayscale camera (Point Gray, Flea 3: FL3-U3-13Y3MC) by changing the height of the camera with 3 patterns and the rotation angle of the turntable with 20 patterns. The size of the image is 1,280x1,024 pixels. A 3D mesh model for each target object is measured with the 3D digitizing system (Steinbichler, COMET5).

• Cat: Grayscale images (51.1MB), Ground truth (29.2 MB)
• Dog: Grayscale images (46.6MB), Ground truth (15.2 MB)

The datasets consist of a set of multi-view images and the ground-truth mesh model. The target objects are figurines of a cat and a dog. The images are captured by an RGB-color camera (Panasonic Lumix GF6) with 2,272x1,704 pixels. The images are taken with the camera by changing the height of the camera with 3 patterns for Cat and 2 patterns for Dog and the rotation angle of the turntable with 36 patterns. Hence, Cat and Dog datasets consist of 108 and 72 images, respectively. A 3D mesh model for each target object is measured with the 3D digitizing system (Steinbichler, COMET5), which is the same in the case of grayscale datasets.

• Cat: RGB-color images (273MB), Ground truth (29.2 MB)
• Dog: RGB-color images (165MB), Ground truth (15.2 MB)

S. Sakai, K. Ito, T. Aoki, T. Watanabe, and H. Unten "Phase-based window matching with geometric correction for multi-view stereo," IEICE Transac\ tions on Information and Systems, Vol. E98-D, No. 10, pp. 1818--1828, October 2015.