medo / image-categoriy-classifier Goto Github PK

Implementation of dense sift in 10x10 blocks. So with this approach I actually iterate over each image and calculate dense sift for each 10x10 block. The time it takes for a large image is too long, because I divide each image in 10x10 blocks and for each block dense sift is calculated.

Integrate the following library for feature extraction http://www.vlfeat.org/install-c.html

Please communicate the results here !

A script has been implemented in order to extract needed information regarding what image belongs to what category, a database (*.csv) file is generated for each category that maps numerical label to image location and the script also generates an extra file that maps numerical label values to its according category name.

Usuage:
./script_pascal.py project_folder
where project_folder is the location of the image-categoriy-classifier directory

Note I did not modify the code to run on the dataset yet, just made Pascal Dataset similar to our internal code representation.

Update the configuration files for modular run, like only extract features,...

Create linear SVM classifier. with the ability of changing the classifier in a config file.

As we've decided refactoring was needed, I'm done implementing it. @medo

Calculated dense sift for the entire image at once (not dividing it into blocks) whether while creating vocabulary, while training and testing.