CVPR2012_Experiments

Running the CVPR 2012 Experiments

First, make sure you have the Proximity Forest and required dependencies installed. See the Installation page for details.

The instructions here are divided into data sets: KTH Actions, Cambridge Gestures, and UCF Sports. For all instructions <proximity_forest_dir> is the directory where the top-level folder in the project source code is located. It should have a single subdirectory, src.</proximity_forest_dir>

KTH Actions

• Change to the <proximity_forest_dir>/src/evaluation directory</proximity_forest_dir>

• Run ipython

• In the ipython interpreter, load up the required code by entering the following:

  run CVPR2012/KTH_Experiments.py

• Load the pre-computed tracklets data file:

  KDat = unPickleData()

• Run one of the experiments in KTH_Experiments.py, for example:

  rc = computeAvgConfusion(KDat, N=27, Tau=21, NumTrials=5)

Cambridge Gestures

• Change to the <proximity_forest_dir>/src/evaluation directory</proximity_forest_dir>

• Run ipython

• In the ipython interpreter, load up the required code by entering the following:

  run CVPR2012/CG_Experiments.py

• Load the pre-computed tracklets data file:

  GDat = unPickleData()

• Run one of the experiments in CG_Experiments.py, for example:

  rc = computeAvgErrBySet(GDat, N=27, Tau=21, NumTrials=10)

UCF Sports

• Change to the <proximity_forest_dir>/src/evaluation directory</proximity_forest_dir>

• Run ipython

• In the ipython interpreter, load up the required code by entering the following:

  run CVPR2012/UCF_Experiments.py

• Load the pre-computed tracklets data file:

  UDat = unPickleData()

• Run the LeaveOneOut classification protocol:

  rc = LeaveOneOut(UDat, N=27, Tau=21)