CVC-13: Multimodal Stereo Dataset

Date:28 May, 2016

CVC-13: Multimodal Stereo Dataset

    CVC-Multimodal Stereo Dataset consists of three subsets of multimodal stereo images (visible and infrared), each one containing a different family of scenarios: Roads, Facades, and Smooth Surfaces. Each subset contains images taken in different conditions and days. For every couple of multimodal stereo images a real or a synthetic depth map, to be used as a ground truth, is given. Additionally, the corresponding calibration sets are also provided.
    The images have been recorded using a multimodal stereo rig with the following characteristics:

    • Color camera: Sony ICX084 Bayer pattern CCD, 6mm focal length lens, 640×480 image resolution (Bumblebee 2 stereo).
    • Infrared camera: PathFindIR Flir camera, 19mm focal length lens (36 degree HFOV), 534×426 image resolution.

    All the images are provided in lossless PPM and BMP format.

    References to this database should be made to the following publication:

    • Barrera F., Lumbreras F. and Sappa A., “Multimodal Stereo Vision System: 3D Data Extraction and Algorithm Evaluation”IEEE Journal of Selected Topics in Signal Processing, Vol. 6, No. 5, September 2012, pp. 437-446.



    Sets of Calibration Pattern
    Name Visible Infrared
    Calibration Dataset (148MB)

    Our calibration data, valid range till 20m.

    Evaluation and validation

    Sets of Images
    Name Visible Infrared Groundtruth
    Roads (11.8MB)
    Facades (4.4MB)
    Smooth Surfaces (129MB)
    OSU Color-Thermal Database*** +/- 2px

    ***This benchmark belongs to OTCBVS, see the link for more datails.