This is just a simple implementation of a classifier that uses Histogram of Oriented Gradients descriptors. Just wanted to code this myself to get the hang of it.
Labeled training samples that were used for testing can be found here.
Contains the descriptor extraction function extract_hog(img), where img is either a path string or an image data array.
Contains the training function train(gamma_, C_, pool), where gamma_ and C_ are the according SVC parameters and pool is a multiprocessing pool object. The function extracts descriptors of the given dataset: samples_from_class and tests_from_class samples per existing label for training and testing accordingly. Provided image files should be stored in the directory specified by path variable. The directory should also include a comma-seperated values file gt.csv of "file, label" rows sorted in ascending order by label. After extracting descriptors and training an SVC, the function computes and displays the accuracy score achieved by the resulting machine over the testing set. The function then provides the means to save the machine in a separate file.
Contains the fit_and_classify(machine_file, samples) function, where machine_file is the path to the SVC-machine file used for evaluation and samples is the data set to be evaluated.
Contains the output of a sequence that evaluated train(gamma_, C_, pool) over a logarithmic grid of C and gamma parameters with samples_from_class = 195 and tests_from_class = 25.