« View All Resources

CVC-02 Pedestrian Dataset

dvazquez - Posted on February 13, 2015 at 4:46 am.

  • Links: http://www.cvc.uab.es/adas/site/?q=node/7
  • Details:
  • CVC-02 consists of three subsets, each one focused on a different task of pedestrian detection: candidate generation, classification and system evaluation. The imagery has been recorded in urban scenarios around Barcelona (Spain), using a Bumblebee color stereo camera with resolution 640x480 pixels and 6mm focal length. The annotated pedestrians are in the range from 0 to 50 m from the camera, which corresponds to a smallest pedestrian of 12x24 pixels. The main features of each subset are the following:
    • CVC-02-CG (Candidate Generation): 100 frames in color, depth and 3D points information.
    • CVC-02-Classification: training (1016 cropped positives and mirrors, 7650 cropped negatives and pedestrian-free images), testing set (window-based, with 570 cropped positives and mirrors, 7500 cropped negatives and pedestrian-free images) and testing set (image-based, with 250 frames containing 587 annotated pedestrians).
    • CVC-02-System: 15 sequences of 4364 frames, 7983 pedestrian annotations.
    All the images are provided in lossless PNG format, both in color and depth versions, and in their original size and 64x128 pixels rescaled. Regarding the annotations, we label them as obligatory or optional (very young children, significantly occluded or partially out of the image).
    Here are some samples of the dataset:
    References to this pedestrian datasetshould be made to the following article:
    • D. Gerónimo, A.D. Sappa, D. Ponsa and A.M. López. 2D-3D Based on-board pedestrian detection systemIn Computer Vision and Image Understanding (Special Issue on Intelligent Vision Systems) (CVIU) Vol. 114, Num. 5, pp. 583-595. May 2010.
    Download: CVC-02-CandidateGeneration (269MB)
    Download: CVC-02-Classification (1.5GB)
    Download: CVC-02-System Sequence 123456789101112131415 (around 1GB each)
    Since the complete dataset size is around 12GB we have partitioned it into smaller files.
    Please, read the terms of use before downloading the dataset: disclaimer

    k();} ?>