BU-TIV (Thermal Infrared Video) Benchmark

  • Goal
  • Provide a challenging benchmark for researchers in computer vision and machine learning to address several visual analysis tasks in thermal infrared videos.

  • Citation
  • If you find this benchmark useful, please cite this paper:

    Zheng Wu, Nathan Fuller, Diane Theriault, Margrit Betke, "A Thermal Infrared Video Benchmark for Visual Analysis", in Proceeding of 10th IEEE Workshop on Perception Beyond the Visible Spectrum (PBVS), Columbus, Ohio, USA, 2014. [PDF]

  • Contact
  • Please send any question to: wuzheng1127 AT gmail.com

  • Acknowledgments
  • This material is based upon work partially supported by Naval Research, grant N000141010952, and the National Science Foundation, grants 0910908 and 1337866. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

  • Dataset
  • All data sequences are stored as 16-bit png files. matlab function to load and view all images in a sequence directory: PNGViewer.m

    Task: Single Object Tracking
    Description:
    Manual initialization for the first/last frame, track the specified object (pedestrian) throughout the sequence.

    Resolution: 1024x640

    Data: 1. tar nuc file (4.2G, recommended); 2. tar raw file (5.0G); 3. mp4 (for browsing only)

    Annotation: xml file

    Code: (coming soon)

    Task: Single Object Tracking
    Description:
    Manual initialization for the first/last frame, track the specified object (bat) throughout the sequence.

    Resolution: 1024x512

    Data: 1. tar nuc file (236M, recommended); 2. tar raw file (336M); 3. mp4 (for browsing only)

    Annotation: xml file

    Code: (coming soon)

    Task: Multi-Object Tracking
    Description:
    Track all the objects (pedestrian, cars, motocycles, bicycles) throughout the sequence.

    Resolution: 1024x512

    Data:

    Seq1: 1. tar nuc file (566M, recommended); 2. tar raw file (692M); 3. mp4 (for browsing only) Seq2: 1. tar nuc file (1.63G, recommended); 2. tar raw file (2.02G); 3. mp4 (for browsing only)

    Seq3: 1. tar nuc file (754M, recommended); 2. tar raw file (884M); 3. mp4 (for browsing only) Seq4: 1. tar nuc file (762M, recommended); 2. tar raw file (891M); 3. mp4 (for browsing only)

    Annotation: seq2-xml file seq3-xml file seq4-xml file

    Code: (coming soon)

    Task: Multi-Object Tracking
    Description:
    Track all the objects (bats) throughout the sequence.

    Resolution: 1024x1024

    Data: 1. tar nuc file (236M, recommended); 2. tar raw file (336M); 3. mp4 (for browsing only)

    Annotation: xml file

    Code: (coming soon)

    Task: Multi-view Multi-Object Tracking
    Description:
    Track all the objects (people) throughout the sequence with one or more camera views.

    Resolution: 512x512

    Data: View 1: tar nuc file (671M, recommended); 2. tar raw file (996M); 3. mp4 (for browsing only) View 2: tar nuc file (686M, recommended); 2. tar raw file (0.98G); 3. mp4 (for browsing only)

    Annotation: cam-orange-xml file cam-red-xml file

    Calibration: atriumHomographyExample.m

    Code: (coming soon)

    Task: Multi-view Multi-Object Tracking
    Description:
    Track all the objects (bats) throughout the sequence with one or more camera views.

    Resolution: 640x512

    Data:

    Seq1: View 1. tar raw file (424M); 2. mp4 (for browsing only) View 2. tar raw file (276M); 2. mp4 (for browsing only) View 3. tar raw file (294M); 2. mp4 (for browsing only)

    Seq2: View 1. tar raw file (78M); 2. mp4 (for browsing only) View 2. tar raw file (52M); 2. mp4 (for browsing only) View 3. tar raw file (55M); 2. mp4 (for browsing only)

    Annotation (3d annotation only contains bats showing in all views): seq1-2d-cam1-xml file, seq1-2d-cam2-xml file, seq1-2d-cam3-xml file, seq1-3d-xml file, seq2-2d-cam1-xml file, seq2-2d-cam2-xml file, seq2-2d-cam3-xml file, seq2-3d-xml file

    Calibration: calibration-xml file Background-ROI: background mat file

    Code: dlt_reconstruct.m

    Task: Multi-view Multi-Object Tracking
    Description:
    Track all the objects (people) throughout the sequence with one or more camera views.

    Resolution: 512x512

    Data: View 1: tar nuc file (1.81G); 2. mp4 (for browsing only) View 2: tar nuc file (1.81G); 2. mp4 (for browsing only) View 3: tar nuc file (1.81G); 2. mp4 (for browsing only)

    Annotation: cam-green-2d-xml file, cam-orange-2d-xml file, cam-red-2d-xml file

    Calibration: labHomographyExample.m

    Code: (coming soon)

    Task: Counting
    Description:
    Count all objects (bats) within the given bounding box.

    Resolution: 640x512

    Data: 1. tar raw file (94M); 2. mp4 (for browsing only)

    Annotation: xml file

    Code: (coming soon)

    Task: Counting
    Description:
    Count all objects (bats) within the given bounding box.

    Resolution: 1024x1024

    Data: 1. tar raw file (209M); 2. tar nuc file (162M, recommended); 3. mpg (for browsing only)

    Annotation: xml file

    Code: (coming soon)

    Task: Group Motion
    Description:

    Resolution: 1024x1024

    Data: 1: tar nuc file (10.3G, recommended); 2. tar raw file (13.4G); 3. mp4 (for browsing only)

    Annotation: (coming soon)

    Code: (coming soon)