2 Classical methods

Tutorial	Description	Link
Introduction: VLAD and Fisher Vectors	Train classical embedders VLAD and Fisher Vector on RootSIFT features, previously state-of-the-art methods for image retrieval.	VLAD and Fisher Vectors
VLAD and Fisher Vector with Deep CNN features	Train classical embedders VLAD and Fisher Vector on deep features extracted from activation maps of a pretrained CNN.	VLAD and Fisher Vectors with deep Features
Embedder pipeline	Combine a VLAD and a Fisher Vector embedder on deep CNN features into a Pipeline and compare images with the concatenated vectors.	Pipeline
Custom RootSIFT feature extractor	Implement RootSIFT on top of the SIFT implementation of OpenCV and inherit FeatureExtractorBase instead of the built-in RootSIFT module, and train a VLAD embedder with it.	Custom Feature Extractor with RootSIFT
Custom ORB feature extractor	Write your own ORB feature extractor by inheriting from FeatureExtractorBase, train a VLAD embedder with it and compare two images.	Custom Feature Extractor with ORB