This is a personal, unofficial implementation of bags of colors for image descriptors. It is based on the following paper:
Christian Wenger, Matthijs Douze, Hervé Jégou, "Bag-of-colors for improved image search". Online: https://dl.acm.org/citation.cfm?id=2072298.2072034
Serve the notebook file with Jupyter. Image data sets need to be obtained and prepared separately, and some constants may have to be updated accordingly.
In order to prepare the data sets, create two directories, training_data and testing_data, each containing nothing but standard image files (e.g. PNG). A substantial amount of images is recommended (>10k).
Python 3 is required. Aside from the given Python dependencies in requirements.txt, the Python bindings for Faiss must also be installed in the system. Please follow the instructions in the official repository on building Faiss for Python. GPU support is recommended, but the code can be easily adjusted to function without it.
MIT