howoongjun / localfeature

EventPointNet: Keypoint detection method with neuromorphic vision data

Local feature evaluation framework

This repository provides a framework for local feature evaluation

How to run

Argument for the method is as follows

Argument	Abbreviation	Description	Option
--model	--m	Model select	eventpointnet, superpoint, sift, orb
--channel	--c	Select channel of the image	default = 3
--mode	--o	Mode Select	makedb, query, match, train
--query	--q	Image query file path(Only for query and match mode)
--match	--a	Image match file path(Only for match mode)
--thresh	--t	Threshold value for keypoint number	default = 3000
--db	--d	DB path for training(Only for train mode)
--ransac	--r	RANSAC threshold value

With the arguments above, you can run the code by executing run.py. Argument mode and model are mandatory for the execution.

python run.py --mode <MODE> --model <MODEL> 

Dependencies

python 3.8.5
pytorch 1.7.1
opencv 4.5.2
scikit-image 0.18.1
numpy 1.19.2

Aside from above dependencies, docker container for this repository is provided here

Evaluation