Caffe support different file formats. LMDB and HDF5 are the ones which are of great interest. Caffe users usually prefer LMDB due to speed and memory matters. However there is a necessity to transform usual file formats to the formats adaptable by the Caffe framework. Numpy is a famous package by python which makes it very simple to create ".npy" file from features which have to be used for machine learning and classification purposes. This project is aimed to transform a pack of numpy files to a single LMDB file.
This file creates an LMDB file from numpy files. The application is face varification in which the LMDB file will be created from pairs of images. The genuine pairs get label "1" and the imposter pairs get label "0". The genuine files contains "gen" in their file name. For this process, the numpy files are HOG features but they basically acn be anything saved in numpy files. This file do the following: 1 - load all the ".npy" data from a folder and create a big numpy array 2 - Create an LMDB file from that numpy array