dbotsw

C++ implementation of “Dense Bag-of-Temporal-SIFT-Words”

Requirements

dbotsw needs the following external librairies (alphabetic order):

• Boost
• LibSVM
• OpenCV

Compilation

make clean && make dbotsw

Example

To launch the example:

• go to dbotsw/
• compile according to the instruction
• execute with: ./bin/dbotsw —p example/param/1 —f example/data/ —o example/output/
• or execute with: /bin/dbotsw —p example/param/1 —t example/data/data_TRAIN —x example/data/data_TEST —o example/output/

Options

Execution options (alphabetic order):

• —f : Folder ‘xx’ containing the data under the format xx/xx_TRAIN + xx/xx_TEST
• —o : Output folder
• —p : Parameter file
• —t / —x : Training & Testing files

Parameter file

An example of good parameter file is given in example/param/1.

Possible parameters that can be included in the parameter file :
sz_block, n_blocks, k (k-means), c_svm, n_sc, k_fold, dense, norm_bow, norm_bow_ssr…

Extraction of feature vectors

Main program to extract feature vectors (and work directly from them) is available.

make clean && make get_feature_vectors

Require options —f and —o