ROUGH DRAFT authorea.com/39605
Main Data History
Export
Show Index Toggle 0 comments
  •  Quick Edit
  • Different SVM Kernel experiment

    The experiment is performed with UCF Sport Action Dataset. Each feature vector is normalized visual word frequency histogram of local features extracted from supervoxel region in the video. The split consists of training \(3886\), testing \(2277\) instances, respectively with \(20k\) dimension.

    • Linear Kernel
      - MAP: \(31.91\%\) ( Runtime* = \(2.5\) mins )

    • Intersection Kernel
      - MAP: \(35.65\%\) ( Runtime* = \(3.3\) mins )

    • Chi-square Kernel
      - MAP: \(39.03\%\) ( Runtime* = \(2.4\) mins )

    • Jenson-Shannon Kernel
      - MAP: \(39.50\%\) ( Runtime* = \(3.0\) mins )

    Runtime* : Test & Training time

    UCF vertex 10

    1. Kernel Type: Linear

    mAP: 31.91 % Time complexity: 2.5 mins

    2. Kernel Type: Intersection

    mAP: 35.65 % Time complexity: 3.3 mins

    3. Kernel Type: Chi-Square

    mAP: 39.03 % Time complexity: 2.4 mins

    4. Kernel Type: Jenson-Shannon

    mAP: 39.50 % Time complexity: 3.0 mins

    UCF vertex 10 bg alfa 0.5

    1. Kernel Type: Linear

    mAP: 56.96 % Time complexity: 22.3 mins

    2. Kernel Type: Intersection

    mAP: 58.38 % Time complexity: 20.3 mins

    3. Kernel Type: Chi-Square

    mAP: 63.08 % Time complexity: 14.8 mins

    4. Kernel Type: Jenson-Shannon

    mAP: 63.53 % Time complexity: 78.7 mins

    UCF vertex 10 bg alfa 0.75

    1. Kernel Type: Linear

    mAP: 65.29 % Time complexity: 15.1 mins

    2. Kernel Type: Intersection

    mAP: 60.87 % Time complexity: 16.8 mins

    3. Kernel Type: Chi-Square

    mAP: 65.96 % Time complexity: 11.2 mins

    \(mAP\) performance with different types of kernel functions vs background histogram weight values, \(\alpha_{bg}\)

    Kernel Type \(\alpha_{bg} = 0\) \(\alpha_{bg} = 0.5\) \(\alpha_{bg} = 0.75\)
    \(Linear\) 31.91 % 56.69 % 65.29 %
    \(Intersection\) 35.65 % 58.98 % 60.89 %
    \(Chi-Square\) 39.03 % 63.08 % 65.96 %
    \(Jenson-Shannon\) 39.50 % 63.53 %

    The runtime evaluation (in mins) with different types of kernel functions vs background histogram weight values, \(\alpha_{bg}\)

    Kernel Type \(\alpha_{bg} = 0\) \(\alpha_{bg} = 0.5\) \(\alpha_{bg} = 0.75\)
    \(Linear\) 2.5 22.3 15.1
    \(Intersection\) 3.3 20.3 16.8
    \(Chi-Square\) 2.4 14.8 11.2
    \(Jenson-Shannon\) 3.0 78.7